[
{
"venue": "ECIR",
"title": "Mixed Reality Ultrasound-Guided Mini-ECIRS with Apple Vision Pro™ - First Case Report",
"authors": [
"Roberto Montoro Neto",
"Fábio C. Vicentini",
"Ricardo T.S. Ugino",
"Alexandre Danilovic",
"Giovanni Scala Marchini",
"Fábio César Miranda Torricelli",
"Carlos Batagello",
"Anderson B. Pellanda",
"Alexandre Sérgio Silva",
"William Carlos Nahas",
"Eduardo Mazzucchi"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1590/s1677-5538.ibju.2024.0610",
"source": "openalex",
"doi": "https://doi.org/10.1590/s1677-5538.ibju.2024.0610",
"abstract": "INTRODUCTION: Some endourological surgeries require multiple screens to perform combined procedures, which can present ergonomic challenges (1, 2). Apple Vision Pro (AVP) is a spatial computing device developed by Apple that incorporates virtual reality (VR) for life-like simulations, realistic medical scenarios, interactive anatomical models, and augmented reality (AR) technologies (3). In health care, VR is used for pain management, physical therapy, psychological therapy, and surgical simulations, providing a controlled and safe environment for both patients and healthcare professionals (4). OBJECTIVE: To demonstrate the step-by-step technique of the Mini-Endoscopic Combined Intra-Renal Surgery (Mini-ECIRS) procedure guided by ultrasound and using mixed reality technology with the Apple Vision Pro (multiscreen and 3D reconstruction). To the best of our knowledge, this is the first report of this procedure being performed with AVP assistance. PATIENT AND METHODS: We present the case of a 40-year-old female with a history of right lumbar pain for one year. A CT scan revealed a proximal ureteral stone (20mm) and a lower pole stone (14mm) on the right side, with a Guys's Score grade 2 4. In this case, we opted for Ultrasound-Guided Mini-ECIRS (5, 6). This choice allowed for precise puncture and dilation, ensuring effective treatment and minimal invasiveness, assisted by the Apple Vision Pro. This device is equipped with eight external cameras that capture the real world at a resolution of 4K, enhancing the surgeon's experience with unparalleled efficiency and ease of mixed reality. This advanced imaging allows for precise visualization and integration of digital elements into the physical environment, significantly improving the accuracy and effectiveness of surgical procedures. During this procedure, the multitude of equipment in the operating room often obstructs the view of the physical monitors, including ultrasound. However, this technology addresses these challenges by offering enhanced ergonomics, efficiency, and safety to the surgeon. By providing seamless integration of digital overlays and real-world visuals, it ensures that crucial information is always within the surgeon's line of sight, thereby improving operational precision and overall outcomes. The surgeon had no previous contact with the AVP and was assisted by an AVP expert urologist throughout the procedure. RESULTS: The procedure was performed in the Barts flank-free position. Initially, ureterolithotomy was performed using holmium laser. After the dusting phase, an ultrasound-guided renal puncture was performed using a virtual screen, providing enhanced comfort and ergonomics for the surgeon. Throughout the procedure, the surgeon had simultaneous access to both screens (nephroscope and flexible ureteroscope), facilitating efficient location of any residual stones. The AVP functioned effectively, displaying multiple screens within its own interface, improving ergonomics during surgery and maintaining safety throughout the procedure. The surgery was performed uneventfully in 2 hours, and the patient was rendered stone-free on CT and was discharged on the first postoperative day. CONCLUSION: Apple Vision Pro provides multiscreen and 3D reconstruction capabilities, ensuring a comfortable, safe, and easily replicable procedure. Its advanced technology may be particularly beneficial for surgeries, such as Mini-ECIRS, which require simultaneous screens."
},
{
"venue": "ECIR",
"title": "Vacuum-assisted mini-ECIRS for calyceal diverticular stones in a recipient of a kidney transplant: A case report",
"authors": [
"Ponthakorn Srithongsongsaeng",
"Kun Sirisopana",
"Surawach Piyawannarat",
"Yada Phengsalae",
"Premsant Sangkum",
"Wisoot Kongchareonsombat",
"Chinnakhet Ketsuwan"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.eucr.2025.103079",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.eucr.2025.103079",
"abstract": "Urolithiasis in renal allografts can cause serious complications, particularly when involving calyceal diverticula due to altered anatomy. We report the case of a 44-year-old female recipient of a kidney transplant who presented with recurrent urinary tract infections and elevated serum creatinine levels. Imaging revealed a calyceal diverticular stone. After a failed retrograde approach, she underwent vacuum-assisted mini-endoscopic combined intrarenal surgery (ECIRS). A flexible ureteroscope enabled precise puncture and direct visualization for diverticular neck incision and fulguration. The procedure was successful with no complications. Vacuum-assisted mini-ECIRS is a safe and effective option for complex stone cases in recipients of transplants."
},
{
"venue": "ECIR",
"title": "Endoscopic Synergy: Endoscopic Combined Intrarenal Surgery (ECIRS)-Guided “Cut-to-Light” Holmium Laser Retrograde Endoureterotomy in Ureteral Stricture Management",
"authors": [
"Suryaram Aravind",
"Punith Jain R",
"Velmurugan Palaniyandi",
"Hariharasudhan Sekar",
"Sriram Krishnamoorthy"
],
"year": 2025,
"pdf_url": "https://assets.cureus.com/uploads/case_report/pdf/344022/20250227-237755-7f352i.pdf",
"source": "openalex",
"doi": "https://doi.org/10.7759/cureus.79758",
"abstract": "This is a case report of a 50-year-old diabetic woman with chronic obstructive pulmonary disease who had an incidentally detected right-sided ureteral stricture. Ureteral strictures are a serious condition that may sometimes progress silently, ultimately resulting in ipsilateral renal function loss. Proper timing and appropriate treatment are essential to preserve renal function and prevent further complications. The management of long-segment strictures ranges from open repair to laparoscopic, robotic, and interventional techniques. This manuscript presents a case of a long-segment ureteral stricture successfully treated using a minimally invasive endourological technique, preserving normal urinary flow. This is the first study to report a successful synergistic interventional approach for the management of complex ureteral strictures."
},
{
"venue": "ECIR",
"title": "ACERVO RESGATE DO PROJETO ECIRS",
"authors": [
"Roberto Radünz",
"Anthony Beux Tessari"
],
"year": 2025,
"pdf_url": "https://revistas.ufg.br/revistaufg/article/download/77817/42499",
"source": "openalex",
"doi": "https://doi.org/10.5216/revufg.v24.77817",
"abstract": "Na história da preservação do patrimônio cultural no Brasil, o estado assumiu um papel preponderante, por meio do Instituto do Patrimônio Histórico e Artístico Nacional (Iphan). Não obstante, iniciativas também ocorreram em outros âmbitos, como é o caso da experiência de registro do patrimônio a partir de uma instituição privada de ensino superior no Sul do Brasil. Neste texto, expõe-se o trabalho e o acervo constituído pelo projeto Ecirs, vinculado ao Instituto Memória Histórica e Cultural (IMHC) da Universidade de Caxias do Sul (UCS). Este projeto, com origem na década de 1970, produziu extensa documentação alusiva à cultura de imigração italiana no estado do Rio Grande do Sul, e atuou em atividades de salvamento e preservação do patrimônio cultural de áreas atingidas pela construção de reservatórios (barragens) de usinas hidrelétricas, nos estados do Rio Grande do Sul e de Santa Catarina. Na década de 1990, o projeto alcançou reconhecimento nacional, sendo condecorado pelo Iphan com o prêmio Rodrigo Melo Franco de Andrade. A fim de apresentar o acervo constituído, contextualiza-se a sua origem, os mecanismos e as lógicas de sua produção, e oferece-se um recorte relacionado ao tema “festas populares”, a partir da documentação oriunda dos projetos das barragens."
},
{
"venue": "ECIR",
"title": "[Effect of Body Mass Index on Outcomes of Mini-ECIRS for Renal Stone].",
"authors": [
"Tetsuo Fukuda",
"Hiroki Ito",
"Takahiko Watanabe",
"Tadashi Tabei",
"Fukashi Yamamichi",
"Takaaki Inoue",
"Kazuki Kobayashi",
"Junichi Matsuzaki"
],
"year": 2025,
"pdf_url": "http://hdl.handle.net/2433/293310",
"source": "openalex",
"doi": "https://doi.org/10.14989/actauroljap_71_3_71",
"abstract": "We retrospectively compared treatment outcomes and complications based on body mass index (BMI) in patients with renal stones treated with mini-endoscopic combined intrarenal surgery (ECIRS) using percutaneous tracts 20 Fr or smaller. Among 1,432 patients who had ECIRS performed at multiple registered facilities between January 2015 and December 2022, 870 patients with renal stones who underwent mini-ECIRS were included after excluding those with anatomical anomalies or incomplete clinical data. The patients were divided into two groups : BMI ≥30 (Group A) and BMI ＜30 (Group B). The treatment outcomes and complications were compared between the two groups. One month postoperatively, plain computed tomography (CT) and kidney ureter bladder radiography (KUB) were performed to assess stone fragmentation and hydronephrosis. According to postoperative imaging, stone-free was defined as residual fragments 4 mm or less on KUB and 2 mm or less on CT. Of the 870 patients, 86 were in Group A and 784 in Group B. The median (interquartile range) cumulative stone diameter was 33.8 (26.35-50.75) mm in Group A, and 32 (24-47) mm in Group B, respectively. The median operative time was 122.5 (92.25- 166.75) min in Group A and 114.5 (89.75-156) min in Group B. The mean and median (interquartile range) postoperative hospital stay were 5.9±2.5 days and 5 (4-7) days in Group A, and 5.4±3.3 days and 5 (4-6) days in Group B. Stone-free rates were 77.9% (67 cases) by KUB and 61.6% (53 cases) by CT in Group A, and 76.1% (597 cases) by KUB and 58.0% (455 cases) by CT in Group B. The incidence of postoperative fever (≥38.0°C) was 38.4% (33 cases) in Group A and 31.8% (249 cases) in Group B, while septic shock occurred in 2.3% (2 cases) of Group A and 2.6% (20 cases) of Group B. A statistically significant difference (p＜0.05) was found in the postoperative hospital stay between the two groups, but no significant differences were observed in the stone-free rates or complication rates. Mini-ECIRS using percutaneous tracts of 20Fr or smaller for renal stones showed no significant difference in SFR and complications between patients with a BMI ≥30 and those with a BMI ＜30."
},
{
"venue": "ECIR",
"title": "KEIR @ ECIR 2025: The Second Workshop on Knowledge-Enhanced Information Retrieval",
"authors": [
"Zihan Wang",
"Jinyuan Fang",
"Giacomo Frisoni",
"Zhuyun Dai",
"Zaiqiao Meng",
"Gianluca Moro",
"Emine Yılmaz"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2501.11499",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2501.11499",
"abstract": "Pretrained language models (PLMs) like BERT and GPT-4 have become the foundation for modern information retrieval (IR) systems. However, existing PLM-based IR models primarily rely on the knowledge learned during training for prediction, limiting their ability to access and incorporate external, up-to-date, or domain-specific information. Therefore, current information retrieval systems struggle with semantic nuances, context relevance, and domain-specific issues. To address these challenges, we propose the second Knowledge-Enhanced Information Retrieval workshop (KEIR @ ECIR 2025) as a platform to discuss innovative approaches that integrate external knowledge, aiming to enhance the effectiveness of information retrieval in a rapidly evolving technological landscape. The goal of this workshop is to bring together researchers from academia and industry to discuss various aspects of knowledge-enhanced information retrieval."
},
{
"venue": "ECIR",
"title": "Mini-Endoscopic Combined Intrarenal Surgery in Patients with Poor Performance Status: A Retrospective Analysis of Postoperative Fever in Over 1000 Cases",
"authors": [
"Tadashi Tabei",
"Hiroki Ito",
"Takaaki Inoue",
"Takahiko Watanab",
"Tetsuo Fukuda",
"Fukashi Yamamichi",
"Yosuke Shibata",
"Junichi Matsuzaki",
"Kazuki Kobayashi"
],
"year": 2025,
"pdf_url": "https://doi.org/10.5152/tud.2025.25013",
"source": "openalex",
"doi": "https://doi.org/10.5152/tud.2025.25013",
"abstract": "Objective: To assess the safety of mini-endoscopic combined intrarenal surgery (mini-ECIRS) in patients with a poor performance status (PS). Methods: A retrospective analysis was conducted on 1132 patients who underwent mini-ECIRS at 3 hospitals between January 2015 and December 2021. Patients were classified according to their PS (PS0-1 and PS2-4 groups) and compared between the groups in terms of preoperative drainage status, such as ureteral stent or percutaneous nephrostomy (PNS), stone characteristics, surgical outcomes, and postoperative fever. Multivariate logistic regression models were used to identify the predictive factors for postoperative fever in each PS group. Results: Patients in the PS2-4 group were older and had a higher stone burden than those in the PS0-1 group. The stone-free rates and surgical success rate were similar between the PS groups, but PS2-4 patients had higher rates of postoperative fever without preoperative drainage. Stone composition analysis revealed a higher prevalence of infectious stones in the PS2-4 group. In the PS0-1 group, PNS reduced postoperative fever risk (odds ratio (OR): 0.65, 95% CI: 0.48-0.89, P = .01), and history of febrile urinary tract infection, stone burden ! 30 mm, number of involved calyces ! 4, and female sex were independent risk factors. Notably, in the PS2-4 group, PNS remained effective against postoperative fever (OR: 0.24, 95% CI: 0.07-0.80, P = .02), while no other factors were significant. Conclusion: The mini-ECIRS was effective even in PS-poor patients, and they may benefit more from preoperative PNS placement than normal PS cases."
},
{
"venue": "ECIR",
"title": "Simultaneous Mini-ECIRS and low-energy TFL endopyelotomy for recurrent UPJO with pelvic renal calculus: A case report",
"authors": [
"Manapol Rujithamkul",
"Kun Sirisopana",
"Surawach Piyawannarat",
"Yada Phengsalae",
"Premsant Sangkum",
"Wisoot Kongchareonsombat",
"Chinnakhet Ketsuwan"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.eucr.2025.103190",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.eucr.2025.103190",
"abstract": "Recurrent ureteropelvic junction obstruction (UPJO) with renal calculi poses a surgical challenge. Miniature Endoscopic Combined Intrarenal Surgery (mini-ECIRS) with low-wattage Thulium Fiber Laser (TFL) enables simultaneous stone clearance and precise endopyelotomy. We report a case of a 51-year-old woman with recurrent UPJO and renal stones treated using mini-ECIRS and TFL-assisted retrograde endopyelotomy. Dual endoscopic access allowed effective stone fragmentation and accurate UPJ incision. Operative time was 75 min, with no complications. This case supports mini-ECIRS with low-energy TFL as a safe, single-session treatment for recurrent UPJO with renal stones."
},
{
"venue": "ECIR",
"title": "Complex staghorn calculus – an enhanced Endoscopic Combined Intra-Renal Surgery (ECIRS) approach using mini-PCNL and suction sheaths",
"authors": [
"Shravankrishna Ananthapadmanabhan",
"John Huynh",
"Henry Wang",
"Nicholas Mehan",
"Michael Myint",
"Nicola Jeffery",
"Celalettin Varol",
"Bertram Canagasingham",
"Jonathan Kam",
"Mohamed Khadra",
"Isaac Thangasamy",
"Raymond Ko"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.urolvj.2025.100351",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.urolvj.2025.100351",
"abstract": "To demonstrate the utility of 3D Modelling to help guide Percutaneous Nephrolithotomy (PCNL) puncture in a complex case of staghorn calculi with a limited window for safe access. Additionally, we demonstrate the efficacy of Endoscopic Combined Intra-Renal Surgery (ECIRS) with use of suction ureteric access sheath and nephrostomy, to perform efficient and total clearance of a complete staghorn calculi. Following detailed discussion and obtainment of informed consent, a complete staghorn calculus was treated with ECIRS. Owing to overlying anatomy including lung, bowel and spleen, the window for safe puncture was miniscule. Using 3D-modelling constructed with non-contrast CT scans, a safe window and calyx was identified and correlated with intra-operative fluoroscopy. A suction ureteric access sheath (ClearPetra, Guangzhou, China) was inserted prior to prone positioning. Following puncture, a mini-PCNL suction nephrostomy tube (ClearPetra, Guangzhou, China) was inserted to facilitate laser lithotripsy. After stone clearance, an indwelling catheter and ureteric stent was left-in-situ, without need for nephrostomy tube. This study demonstrates the principal surgical steps involved with performing endourological treatment of a complex, complete Staghorn Calculi. Steps include pre-operative planning with 3D-modelling, intra-operative puncture under fluoroscopic and direct vision with use of pyeloscopy (ECIRS), use of ureteric access sheath for prevention of ureteric assessment and prevention stone migration, laser lithotripsy via mini-PCNL, and post-operative follow-up. CT imaging on post-operative day 1 confirmed complete stone clearance. Post-operatively the patient was monitored for 2 days with no complications, and catheter was removed prior to discharge. For complex endourological cases, thorough pre-operative planning is vital to ensure good operative outcomes. This case demonstrates the use of 3D-modelling to plan accurate and safe puncture trajectory within a limited window of safety. Additionally, ECIRS with suction sheaths is utilised to effectively and efficiently clear large stone burden."
},
{
"venue": "ECIR",
"title": "LIR: The First Workshop on Late Interaction and Multi Vector Retrieval @ ECIR 2026",
"authors": [
"Benjamin Clavié",
"Xianming Li",
"Antoine Chaffin",
"Omar Khattab",
"Tom Aarsen",
"Manuel Faysse",
"Jing Li"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2511.00444",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2511.00444",
"abstract": "Late interaction retrieval methods, pioneered by ColBERT, have emerged as a powerful alternative to single-vector neural IR. By leveraging fine-grained, token-level representations, they have been demonstrated to deliver strong generalisation and robustness, particularly in out-of-domain settings. They have recently been shown to be particularly well-suited for novel use cases, such as reasoning-based or cross-modality retrieval. At the same time, these models pose significant challenges of efficiency, usability, and integrations into fully fledged systems; as well as the natural difficulties encountered while researching novel application domains. Recent years have seen rapid advances across many of these areas, but research efforts remain fragmented across communities and frequently exclude practitioners. The purpose of this workshop is to create an environment where all aspects of late interaction can be discussed, with a focus on early research explorations, real-world outcomes, and negative or puzzling results to be freely shared and discussed. The aim of LIR is to provide a highly-interactive environment for researchers from various backgrounds and practitioners to freely discuss their experience, fostering further collaboration."
},
{
"venue": "ECIR",
"title": "The benefits of PCNL combined with ureterorenoscopy compared to isolated PCNL in urolithiasis ( PRECISE ): a multicentre randomised controlled trial protocol",
"authors": [
"Chris A. Suijker",
"Riemer A. Kingma",
"Dirk Bakker",
"Ward Goossens",
"J.G.H. Poerink",
"Antoinette D. I. van Asselt",
"Inge M. van Oort",
"Stijn Roemeling",
"the ROUND‐PRECISE study group"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1111/bju.70008",
"source": "openalex",
"doi": "https://doi.org/10.1111/bju.70008",
"abstract": "BACKGROUND: Percutaneous nephrolithotomy (PCNL) often fails to achieve complete stone clearance with a single procedure. Residual fragments, irrespective of size, are associated with increased stone-related morbidity and a higher likelihood of re-intervention. Endoscopic combined intrarenal surgery (ECIRS), which integrates PCNL with retrograde ureterorenoscopy, may improve stone-free rates and reduce morbidity. However, given the potential for increased risks and greater resource utilisation, a comprehensive randomised controlled trial is warranted. The 'PRECISE' trial is designed to compare ECIRS and PCNL in terms of efficacy, safety, stone-related morbidity, and resource utilisation. STUDY DESIGN: The PRECISE is a prospective, non-blinded, randomised, parallel-group, multicentre interventional trial involving patients eligible for percutaneous kidney stone surgery. Patients will be randomised in a 1:1 ratio to undergo either ECIRS or standard PCNL. ENDPOINTS: The primary endpoint is the Grade A (0 mm) stone-free rate on postoperative computed tomography (CT) at 4 weeks. Secondary radiological endpoints include Grade B (≤2 mm) and Grade C (≤4 mm) stone-free rates at 4 weeks, stone volume reduction, 1-year stone-free rates, and radiation exposure. Additional outcomes include complications, operative duration, stone-related events (SREs), health-related quality of life (HRQoL), cost differences, cost-effectiveness and environmental impact, assessed up to 5 years postoperatively. PATIENTS AND METHODS: Adults (aged ≥18 years) eligible for percutaneous kidney stone surgery, where retrograde access is feasible, will be recruited. The main exclusion criteria comprise papillary calcifications, renal transplant procedures, and (suspected) pregnancy. A total of 350 patients (175 per arm) will be recruited. Follow-up includes CT imaging with clinical consultations at 4 weeks and 1 year to evaluate radiological outcomes, complications, SREs, and HRQoL. Telephone interviews at 3 and 5 years will assess long-term SREs and HRQoL. We aim to retrospectively evaluate cost differences, cost-effectiveness, and environmental impact. TRIAL REGISTRATION: The PRECISE study: an investigation into combined kidney stone surgery, Dutch Trial Registry, NL-009577."
},
{
"venue": "ECIR",
"title": "Understanding the Interplay between LLMs' Utilisation of Parametric and Contextual Knowledge: A Keynote at ECIR 2025",
"authors": [
"Isabelle Augenstein"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2603.09654",
"source": "openalex",
"doi": "https://doi.org/10.1145/3799914.3799918",
"abstract": "Language Models (LMs) acquire parametric knowledge from their training process, embedding it within their weights. The increasing scalability of LMs, however, poses significant challenges for understanding a model's inner workings and further for updating or correcting this embedded knowledge without the significant cost of retraining. Moreover, when using these language models for knowledge-intensive language understanding tasks, LMs have to integrate relevant context, mitigating their inherent weaknesses, such as incomplete or outdated knowledge. Nevertheless, studies indicate that LMs often ignore the provided context as it can be in conflict with the pre-existing LM's memory learned during pre-training. Conflicting knowledge can also already be present in the LM's parameters, termed intra-memory conflict. This underscores the importance of understanding the interplay between how a language model uses its parametric knowledge and the retrieved contextual knowledge. In this talk, I will aim to shed light on this important issue by presenting our research on evaluating the knowledge present in LMs, diagnostic tests that can reveal knowledge conflicts, as well as on understanding the characterists of successfully used contextual knowledge. Date: 8 April 2025."
},
{
"venue": "ECIR",
"title": "Mini-endoscopic combined intrarenal surgery (Mini-ECIRS) in the management of staghorn renal calculi",
"authors": [
"Wong‐Kein Low",
"Timothy Yong Kuei Lim",
"Yong Wei Lim",
"P. Sundaram"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/s2666-1683(25)02007-5",
"source": "openalex",
"doi": "https://doi.org/10.1016/s2666-1683(25)02007-5",
"abstract": ""
},
{
"venue": "ECIR",
"title": "Infectious Complications Following Mini‐Endoscopic Combined Intrarenal Surgery at Japanese Tertiary Institutions",
"authors": [
"Takahiko Watanabe",
"Hiroki Ito",
"Tetsuo Fukuda",
"Fukashi Yamamichi",
"Tadashi Tabei",
"Takaaki Inoue",
"Junichi Matsuzaki",
"Kazuki Kobayashi"
],
"year": 2025,
"pdf_url": "https://onlinelibrary.wiley.com/doi/pdfdirect/10.1111/iju.70037",
"source": "openalex",
"doi": "https://doi.org/10.1111/iju.70037",
"abstract": "PURPOSE: To identify risk factors of infectious complications following mini-endoscopic combined intrarenal surgery (ECIRS) in patients with renal or ureteral stones. METHODS: We retrospectively analyzed consecutive patients with renal or ureteral stones who underwent mini-ECIRSs at three Japanese tertiary institutions between 2015 and 2021. Data were collected and evaluated regarding patient backgrounds, stone characteristics, and postoperative complications. Among the various complications, multivariable logistic regression analysis was performed using preoperative and intraoperative factors for postoperative fever (≥ 38°C) and septic shock to identify independent risk factors. RESULTS: The data of 1432 cases were collected. Finally, 1035 cases of single-session mini-ECIRS were included in the analysis. In infectious complications, postoperative fever and septic shock were observed in 273 and 21 patients (26.4% and 2.0%). A multivariable logistic regression model identified female (p < 0.001), ureteral stones (p < 0.001), preoperative pyuria (p < 0.001), preoperative urinary tract infection (p = 0.045), preoperative percutaneous nephrostomy (p = 0.001), and operation time (p = 0.017) as predictors of postoperative fever. For septic shock, female (p < 0.001) was shown as a risk factor. CONCLUSIONS: To the best of our knowledge, this multicenter cohort study is the largest study investigating infectious complications following mini-ECIRS. Female was a common risk factor for both postoperative fever and septic shock, suggesting that surgeons should pay extra attention to vital signs during the procedure and postoperative infectious complications in mini-ECIRSs for these patients."
},
{
"venue": "ECIR",
"title": "Comparison of the microbiome of bladder urine, upper urinary tract urine, and kidney stones in patients with urolithiasis",
"authors": [
"Joanna Chorbińska",
"Wojciech Krajewski",
"Paweł Karpiński",
"Łukasz Nowak",
"Wojciech Tomczak",
"Jan Łaszkiewicz",
"Katarzyna Pacyga-Prus",
"Sabina Górska",
"Bartosz Małkiewicz",
"Tomasz Szydełko"
],
"year": 2025,
"pdf_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12379818/pdf/CEJU-78-02-0020.pdf",
"source": "openalex",
"doi": "https://doi.org/10.5173/ceju.2025.0020",
"abstract": "Introduction: It is believed that bacteria can be involved in the formation of all types of stones. The aim of study was to assess the urinary microbiome in patients with urolithiasis. Material and methods: The study group included 50 patients qualified for endoscopic treatment of urinary tract stones using: ureteroscopic lithotripsy (URSL), retrograde intrarenal surgery (RIRS), percutaneous nephrolithotripsy (PCNL), endoscopic combined intrarenal surgery (ECIRS). Before the procedure, patients were asked to collect urine and stool for analysis. Urine from the upper urinary tract and stone fragments were collected intraoperatively. The research material was subjected to 16S rRNA sequencing. The chemical composition of stones was assessed using Raman spectroscopy. Results: . Further analysis showed the relative similarity of the urinary bladder and upper urinary tract microbiomes and the dissimilarity of the kidney stone microbiome. A comparison of the upper urinary tract microbiome based on the method of urine collection and a comparison of urinary bladder and upper urinary tract microbiomes based on the presence of a DJ stent prior to the procedure showed no statistically significant differences. Conclusions: The microbiome of stones differs from the microbiome of urine, which may play a role in the pathogenesis of urolithiasis. Bladder urine and upper urinary tract urine microbiomes do not differ. Therefore, bladder urine can replace upper urinary tract urine in microbiome studies."
},
{
"venue": "ECIR",
"title": "KIMERA: From Evaluation-as-a-Service to Evaluation-in-the-Cloud",
"authors": [
"Andrea Pasin",
"Nicola Ferro"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730298",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730298",
"abstract": "Experimental evaluation steers the development of Information Retrieval (IR) systems, and large-scale evaluation campaigns provide the field with a common infrastructure to conduct comparable evaluation exercises. Over the years, tools and platforms have been developed to manage and automate these activities, enhance the reproducibility of conducted experiments and facilitate data sharing. In this context, Evaluation-as-a-Service (EaaS) emerged as an approach to avoid distributing experimental collections, which may contain copyrighted or sensitive data, and instead execute containerised code on that data on remote servers. We propose Kubernetes Infrastructure for Managed Evaluation and Resource Access (KIMERA) as the next step from EaaS into Evaluation-in-the-Cloud (EitC), allowing researchers to directly code and execute their systems through their browsers, requiring only an internet connection. Moreover, recent advancements, such as Large Language Models, or new computing paradigms, such as quantum computers, require external third party services and computational resources. In this respect, KIMERA streamlines and simplifies access to such services on-demand via their APIs. More in detail, KIMERA relies on state-of-the-art containerization and orchestration tools, such as Docker and Kubernetes, to provide a robust, scalable, secure, and fault-tolerant IR evaluation platform. KIMERA monitors and stores all the participants' submissions, accurately keeping track of the resource usage, allowing for evaluating both the efficiency and the effectiveness of the deployed methods. Moreover, all participants can be assigned workspaces sharing the same resources (i.e., CPU and RAM), thus enhancing reproducibility and comparability among systems. Finally, KIMERA has been designed with modularity and extensibility in mind, allowing it to be easily adapted to new use cases and usage scenarios. KIMERA has been developed and adopted in the context of the QuantumCLEF lab, to allow for mixed experiments, comparing approaches running on traditional hardware and on real quantum annealers provided by external companies. KIMERA has also been used as a learning resource to provide Quantum Computing tutorials for IR at major conferences, such as ECIR and SIGIR. The source code of KIMERA is openly available at https://github.com/MjPaxter/KIMERA."
},
{
"venue": "ECIR",
"title": "Reference coverage analysis of OpenAlex compared to Web of Science and Scopus",
"authors": [
"Jack H. Culbert",
"Anne Hobert",
"Najko Jahn",
"Nick Haupka",
"Marion Schmidt",
"Paul Donner",
"Philipp Mayr"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s11192-025-05293-3.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s11192-025-05293-3",
"abstract": "Abstract OpenAlex is a promising open source of scholarly metadata, and competitor to established proprietary sources, such as the Web of Science and Scopus. As OpenAlex provides its data freely and openly, it permits researchers to perform bibliometric studies that can be reproduced in the community without licensing barriers. However, as OpenAlex is a rapidly evolving source and the data contained within is expanding and also quickly changing, the question naturally arises as to the trustworthiness of its data. In this report, we will study the reference coverage and selected metadata within each database and compare them with each other to help address this open question in bibliometrics. In our large-scale study, we demonstrate that, when restricted to a cleaned dataset of 16.8 million recent publications shared by all three databases, OpenAlex has average source reference numbers and internal coverage rates comparable to both Web of Science and Scopus. We further analyse the metadata in OpenAlex, the Web of Science and Scopus by journal, finding a similarity in the distribution of source reference counts in the Web of Science and Scopus as compared to OpenAlex. We also demonstrate that the comparison of other core metadata covered by OpenAlex shows mixed results when broken down by journal, where OpenAlex captures more ORCID identifiers, fewer abstracts and a similar number of Open Access status indicators per article when compared to both the Web of Science and Scopus."
},
{
"venue": "ECIR",
"title": "Theory and Toolkits for User Simulation in the Era of Generative AI: User Modeling, Synthetic Data Generation, and System Evaluation",
"authors": [
"Krisztian Balog",
"Nolwenn Bernard",
"Saber Zerhoudi",
"ChengXiang Zhai"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3731697",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3731697",
"abstract": "Interactive AI systems, including search engines, recommender systems, conversational agents, and generative AI applications, are increasingly central to user experiences. However, rigorously evaluating their performance, training them effectively with interaction data, and modeling user behavior for personalization remain significant challenges, often difficult to address reproducibly and at scale. User simulation, which employs intelligent agents to mimic human interaction patterns, offers a powerful and versatile methodology to tackle these interconnected issues. This half-day tutorial provides a comprehensive overview of modern user simulation techniques for interactive AI systems. We will explore the theoretical foundations and practical applications of simulation for system evaluation, algorithm training, and user modeling, emphasizing the crucial connections between these uses. The tutorial covers key simulation methodologies, with a particular focus on recent advancements leveraging large language models, discussing both the opportunities they present and the open challenges they entail. Crucially, we will also provide practical guidance, highlighting relevant toolkits, libraries, and datasets available to researchers and practitioners."
},
{
"venue": "ECIR",
"title": "Query Performance Prediction Using Relevance Judgments Generated by Large Language Models",
"authors": [
"Chuan Meng",
"Negar Arabzadeh",
"Arian Askari",
"Mohammad Aliannejadi",
"Maarten de Rijke"
],
"year": 2025,
"pdf_url": "https://pure.uva.nl/ws/files/308953815/Query_Performance_Prediction_Using_Relevance_Judgments_Generated_by_Large_Language_Models.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1145/3736402",
"abstract": "Query performance prediction (QPP) aims to estimate the retrieval quality of a search system for a query without human relevance judgments. Previous QPP methods typically return a single scalar value and do not require the predicted values to approximate a specific information retrieval (IR) evaluation measure, leading to certain drawbacks: (i) a single scalar is insufficient to accurately represent different IR evaluation measures, especially when metrics do not highly correlate, and (ii) a single scalar limits the interpretability of QPP methods because solely using a scalar is insufficient to explain QPP results. To address these issues, we propose a QPP framework using automatically gen erated re levance judgments (QPP-GenRE), which decomposes QPP into independent subtasks of predicting the relevance of each item in a ranked list to a given query. This allows us to predict any IR evaluation measure using the generated relevance judgments as pseudo-labels. This also allows us to interpret predicted IR evaluation measures, and identify, track, and rectify errors in generated relevance judgments to improve QPP quality. We predict an item’s relevance by using open source large language models (LLMs) to ensure scientific reproducibility. We face two main challenges: (i) excessive computational costs of judging an entire corpus for predicting a metric considering recall, and (ii) limited performance in prompting open source LLMs in a zero-/few-shot manner. To solve the challenges, we devise an approximation strategy to predict an IR measure considering recall and propose to fine-tune open source LLMs using human-labeled relevance judgments. Experiments on the TREC 2019–2022 deep learning tracks and CAsT-19–20 datasets show that QPP-GenRE achieves state-of-the-art QPP quality for both lexical and neural rankers."
},
{
"venue": "ECIR",
"title": "Knowledge Graph-Guided Retrieval Augmented Generation",
"authors": [
"Xiangrong Zhu",
"Yuexiang Xie",
"Yi Liu",
"Yuanman Li",
"Wei Hu"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.naacl-long.449.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.naacl-long.449",
"abstract": "Xiangrong Zhu, Yuexiang Xie, Yi Liu, Yaliang Li, Wei Hu. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025."
},
{
"venue": "ECIR",
"title": "Modeling cross-platform narrative templates: a temporal knowledge graph approach",
"authors": [
"Ridwan Amure",
"Nitin Agarwal"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s13278-025-01429-8.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s13278-025-01429-8",
"abstract": "Abstract Over the past decade, online social media has grown in size, features, and complexity, providing users with increased satisfaction and prompting many to maintain accounts across multiple platforms. Information actors have also taken advantage of this environment, using cross-platform dynamics to amplify content’s reach and target specific audiences strategically. As these actors will likely continue exploiting social media, we argue that it is crucial to model cross-platform narratives effectively and identify the patterns-or templates defined in this research-they use to propagate different narratives. To address these challenges, we leverage temporal knowledge graphs to model the relationships between cross-platform narratives, extract temporal communities representing macro-narratives, and apply sequential mining to uncover various narrative templates. These templates reveal the patterns various actors use to spread different narratives across various social media platforms. An analysis of 4,817 Instagram posts, 2,560 TikTok posts, 11,134 X posts, and 7,327 YouTube posts, demonstrates the efficacy of this approach in identifying the templates preferred by Pro-Taiwan and Pro-China actors in the Asia–Pacific political landscape. We identified two groups of narrative templates based on confidence and support. Our further analysis uncovers which templates were favored by Pro-Taiwan and Pro-China supporters."
},
{
"venue": "ECIR",
"title": "Clinical predictors of blood loss during mini-endoscopic combined intrarenal surgery",
"authors": [
"Hiroki Ito",
"Tetsuo Fukuda",
"Fukashi Yamamichi",
"Takahiko Watanabe",
"Yosuke Shibata",
"Tadashi Tabei",
"Kazuhide Makiyama",
"Takaaki Inoue",
"Junichi Matsuzaki",
"Kazuki Kobayashi"
],
"year": 2025,
"pdf_url": "https://www.researchsquare.com/article/rs-7958967/latest.pdf",
"source": "openalex",
"doi": "https://doi.org/10.21203/rs.3.rs-7958967/v1",
"abstract": ""
},
{
"venue": "ECIR",
"title": "High-entropy alloyed single-atom Pt for methanol oxidation electrocatalysis",
"authors": [
"Mingda Liu",
"Zhichao Zhang",
"Chenyu Li",
"Sen Jin",
"Kunlei Zhu",
"Shoushan Fan",
"Jia Li",
"Kai Liu"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41467-025-61376-y.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41467-025-61376-y",
"abstract": "at only 2.3 at% Pt and maintains high activity even after operation for 180,000 s. Both experimental and theoretical results reveal that the high-entropy structure induces a synergistic effect, wherein the elements coordinated around single-atom Pt sites effectively remove adsorbed CO from Pt. This mechanism facilitates the key reaction steps of methanol oxidation reaction and avoids CO poisoning. This work presents a high-entropy alloyed single-atom strategy to realize efficient and durable methanol oxidation reaction catalysis with low costs."
},
{
"venue": "ECIR",
"title": "Abstracts",
"authors": [
"The management of encrusted and neglected ureteral stents in renal transplant recipients poses significant clinical challenges",
"particularly when initial stent removal attempts are unsuccessful. This case report highlights the use of Endoscopic Combined Intrarenal Surgery (ECIRS) in addressing this complex issue."
],
"year": 2025,
"pdf_url": "https://onlinelibrary.wiley.com/doi/pdfdirect/10.1111/bju.16629",
"source": "openalex",
"doi": "https://doi.org/10.1111/bju.16629",
"abstract": "This approach enabled simultaneous access to the transplanted renal pelvis and upper ureter, facilitating the fragmentation and removal of the stent.The procedure involved the use of advanced endoscopic tools and techniques to manage the significant encrustation and removal of the stent.The ECIRS procedure successfully removed the encrusted stent and patient experienced a favourable recovery with no immediate postoperative complications.Conclusion: This case underscores the effectiveness of ECIRS in managing complex scenarios involving encrusted and neglected ureteral stents in renal transplant patients.The hybrid approach provided a viable solution when traditional methods failed, offering a minimally invasive alternative to more invasive surgical interventions."
},
{
"venue": "ECIR",
"title": "Enhancing Health Information Retrieval with RAG by prioritizing topical relevance and factual accuracy",
"authors": [
"Rishabh Upadhyay",
"Marco Viviani"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s10791-025-09505-5.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s10791-025-09505-5",
"abstract": "Abstract The exponential surge in online health information, coupled with its increasing use by non-experts, highlights the pressing need for advanced Health Information Retrieval (HIR) models that consider not only topical relevance but also the factual accuracy of the retrieved information, given the potential risks associated with health misinformation. To this aim, this paper introduces a solution driven by Retrieval-Augmented Generation (RAG), which leverages the capabilities of generative Large Language Models (LLMs) to enhance the retrieval of health-related documents grounded in scientific evidence. In particular, we propose a three-stage model: in the first stage, the user’s query is employed to retrieve topically relevant passages with associated references from a knowledge base constituted by scientific literature. In the second stage, these passages, alongside the initial query, are processed by LLMs to generate a contextually relevant rich text (GenText). In the last stage, the documents to be retrieved are evaluated and ranked both from the point of view of topical relevance and factual accuracy by means of their comparison with GenText, either through stance detection or semantic similarity. In addition to calculating factual accuracy, GenText can offer a layer of explainability for it, aiding users in understanding the reasoning behind the retrieval. Experimental evaluation of our model on benchmark datasets and against baseline models demonstrates its effectiveness in enhancing the retrieval of both topically relevant and factually accurate health information, thus presenting a significant step forward in the health misinformation mitigation problem."
},
{
"venue": "ECIR",
"title": "Bridging the Gap: From Ad-hoc to Proactive Search in Conversations",
"authors": [
"Chuan Meng",
"Francesco Tonolini",
"Fengran Mo",
"Νικόλαος Αλέτρας",
"Emine Yilmaz",
"Gabriella Kazai"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729915",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729915",
"abstract": "Proactive search in conversations (PSC) aims to reduce user effort in formulating explicit queries by proactively retrieving useful relevant information given conversational context. Previous work in PSC either directly uses this context as input to off-the-shelf ad-hoc retrievers or further fine-tunes them on PSC data. However, ad-hoc retrievers are pre-trained on short and concise queries, while the PSC input is longer and noisier. This input mismatch between ad-hoc search and PSC limits retrieval quality. While fine-tuning on PSC data helps, its benefits remain constrained by this input gap. In this work, we propose Conv2Query, a novel conversation-to-query framework that adapts ad-hoc retrievers to PSC by bridging the input gap between ad-hoc search and PSC. Conv2Query maps conversational context into ad-hoc queries, which can either be used as input for off-the-shelf ad-hoc retrievers or for further fine-tuning on PSC data. Extensive experiments on two PSC datasets show that Conv2Query significantly improves ad-hoc retrievers' performance, both when used directly and after fine-tuning on PSC."
},
{
"venue": "ECIR",
"title": "Rankers, Judges, and Assistants: Towards Understanding the Interplay of LLMs in Information Retrieval Evaluation",
"authors": [
"Krisztian Balog",
"Donald Metzler",
"Zhen Qin"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730348",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730348",
"abstract": "Large language models (LLMs) are increasingly integral to information retrieval (IR), powering ranking, evaluation, and AI-assisted content creation. This widespread adoption necessitates a critical examination of potential biases arising from the interplay between these LLM-based components. This paper synthesizes existing research and presents novel experiment designs that explore how LLM-based rankers and assistants influence LLM-based judges. We provide the first empirical evidence of LLM judges exhibiting significant bias towards LLM-based rankers. Furthermore, we observe limitations in LLM judges' ability to discern subtle system performance differences. Contrary to some previous findings, our preliminary study does not find evidence of bias against AI-generated content. These results highlight the need for a more holistic view of the LLM-driven information ecosystem. To this end, we offer initial guidelines and a research agenda to ensure the reliable use of LLMs in IR evaluation."
},
{
"venue": "ECIR",
"title": "Role and Efficacy of Percutaneous 4.5‐Fr Semi‐Rigid Ureteroscope While Creating Through‐And‐Through Guidewire Passage During Minimally Invasive Endoscopic Combined Intrarenal Surgery in the Supine Position; Retrospective Study at a Single Institution",
"authors": [
"Takaaki Inoue",
"Shuzo Hamamoto",
"Shinsuke Okada",
"Naoto Tanaka",
"Fukashi Yamamichi",
"Masaichiro Fujita",
"Hideaki Miyake"
],
"year": 2025,
"pdf_url": "https://onlinelibrary.wiley.com/doi/pdfdirect/10.1111/iju.70099",
"source": "openalex",
"doi": "https://doi.org/10.1111/iju.70099",
"abstract": "OBJECTIVE: This study was performed to evaluate the efficacy of using a percutaneous 4.5-Fr semi-rigid ureteroscope (4.5-rURS) and preoperative clinical factors in the performance of the through-and-through guidewire technique (T&TGW) under the guidance of retrograde flexible ureteroscope (fURS)-assisted fluoroscopy during minimally invasive endoscopic combined intrarenal surgery (mini-ECIRS). METHODS: In total, 123 patients who underwent mini-ECIRS with percutaneous 4.5-rURS assistance for large kidney stones in the flank-free modified Valdivia position were retrospectively evaluated. The primary outcome was the achievement rate of T&TGW, defined as percutaneous insertion of a guidewire that passed down the ureter and exited outside the external urethral meatus. The secondary outcome was the rate of percutaneous 4.5-rURS assistance required to successfully perform T&TGW. RESULTS: The stone-free rate was 81.2%, including no residual fragments in 72.3% of patients. T&TGW was successful in 119 (96.6%) patients, which was necessary using retrograde fURS-assisted fluoroscopic guidance in 61 (49.5%) patients, whereas 4.5-rURS assistance was required in 58 (47.1%) patients. Guidewire migration under the renal pelvis mucosa was found in 10 (8.1%) patients. Preoperative factors associated with failed T&TGW using retrograde fURS-assisted fluoroscopic guidance were the puncture site in the lower calyx (odds ratio, 0.125; confidence interval, 0.043-0.364; p = 0.001) and the presence of preoperative hydronephrosis in the renal calyx of the puncture site (odds ratio, 0.261; confidence interval, 0.093-0.730; p = 0.011). CONCLUSION: Percutaneous 4.5-rURS assistance facilitated successful T&TGW in cases of failed access under retrograde fURS-assisted fluoroscopic guidance in supine mini-ECIRS and a useful option to find a migrated guidewire in the wrong direction."
},
{
"venue": "ECIR",
"title": "Non-papillary, prone approach for endoscopic combined intrarenal surgery: our experience",
"authors": [
"Angelis Peteinaris",
"Theodoros Spinos",
"Begoña Ballesta Martínez",
"Arman Tsaturyan",
"Solon Faitatziadis",
"Mohammad M. Obaidat",
"Athanasios Vagionis",
"Fotios Michalopoulos",
"Spyridon POLIZONIS",
"Theofanis Vrettos",
"Evangelos Liatsikos"
],
"year": 2025,
"pdf_url": "https://doi.org/10.23736/s2241-9136.25.00092-1",
"source": "openalex",
"doi": "https://doi.org/10.23736/s2241-9136.25.00092-1",
"abstract": "BACKGROUND: This study was conducted in order to assess the safety and efficacy of non-papillary prone endoscopic combined intrarenal surgery (ECIRS) for patients with urolithiasis.METHODS: The surgeries were performed at a high-volume tertiary center. This retrospective study includes 83 patients who underwent ECIRS between February 2019 and March 2025. The inclusion criteria for ECIRS included complex stones, multiple stones in different calyces, and staghorn calculi. The procedures were conducted by two highly-experienced surgeons. The average patient age was 53.96 years, with a mean stone diameter of 40.33 mm. Nonpapillary puncture was performed and either a 22Fr (mini-percutaneous nephrolithotomy [PCNL]) or 30Fr (standard PCNL) access was implemented.RESULTS: The average operative time was 50.92 minutes, with a final stone-free rate (SFR) of 89.2%. Postoperative complications were observed in 14.5% of the patients, all classified as Grade II, with temporary fever and bleeding being the most common. Comparisons between mini-PCNL (22Fr) and standard PCNL (30Fr) revealed no statistically significant differences in operative time, hemoglobin drop, SFR, or complication rates.CONCLUSIONS: Non-papillary prone ECIRS continues to be a feasible,safe and efficient approach for patients with complex stone disease and anatomical abnormalities."
},
{
"venue": "ECIR",
"title": "ReNeuIR at SIGIR 2025: The Fourth Workshop on Reaching Efficiency in Neural Information Retrieval",
"authors": [
"Sebastian Bruch",
"Maik Fröbe",
"Thomas Hagen",
"Franco Maria Nardini",
"Martin Potthast"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730358",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730358",
"abstract": "Measuring effectiveness and efficiency in information retrieval has a strong empirical background. While modern retrieval systems substantially improve effectiveness, the community has not yet agreed on how to measure efficiency, making it difficult to contrast effectiveness and efficiency fairly. Efficiency-oriented system comparisons are difficult due to factors such as hardware configurations, software versioning, and experimental settings. Efficiency affects users, researchers, and the environment and can be measured in many dimensions beyond time and space, such as resource consumption, water usage, and sample efficiency. Analyzing the efficiency of algorithms and their trade-off with effectiveness requires revisiting and establishing new standards and principles, from defining relevant concepts to designing new measures and guidelines to assess the findings' significance. ReNeuIR's fourth iteration aims to bring the community together to debate these questions and collaboratively test and improve benchmarking frameworks for efficiency based on discussions and collaborations of its previous iterations, including a shared task focused on efficiency and reproducibility."
},
{
"venue": "ECIR",
"title": "Enhancing Recommender Systems: Deep Modality Alignment with Large Multi-Modal Encoders",
"authors": [
"Zixuan Yi",
"Zijun Long",
"Iadh Ounis",
"Craig Macdonald",
"Richard McCreadie"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3718099",
"source": "openalex",
"doi": "https://doi.org/10.1145/3718099",
"abstract": "In recent years, the rapid growth of online multimedia services, such as e-commerce platforms, has necessitated the development of personalised recommendation approaches that can encode diverse content about each item. Indeed, modern multi-modal recommender systems exploit diverse features obtained from raw images and item descriptions to enhance the recommendation performance. However, the existing multi-modal recommender systems primarily depend on the features extracted individually from different media through pre-trained modality-specific encoders, and exhibit only shallow alignments between different modalities, thereby limiting these systems’ ability to capture the underlying relationships between the modalities. In this article, we enhance the deep alignment of large multi-modal encoders to address the shallow alignment of modalities in multi-modal recommender systems. These encoders have previously demonstrated state-of-the-art effectiveness in ranking items across various domains. Specifically, we investigate the use of three state-of-the-art large multi-modal encoders – CLIP (dual-stream), VLMo and BEiT-3 (unified) – for recommendation tasks. We explore their benefits for recommendation through using a range of strategies, including the use of pre-trained and fine-tuned encoders, as well as the evaluation of the end-to-end training of these encoders. We show that pre-trained large multi-modal encoders generate more aligned and effective user/item representations compared with existing modality-specific encoders across four existing multi-modal recommendation datasets. Furthermore, we show that fine-tuning these encoders further improves the recommendation performance, with end-to-end training emerging as the most effective paradigm, significantly outperforming both pre-trained and fine-tuned encoders with an improved recommendation performance. We also demonstrate the effectiveness of large multi-modal encoders in facilitating modality alignment by evaluating the contribution of each modality separately. Finally, we show that the dual-stream approach, specifically CLIP, is the most effective architecture for these large multi-modal encoders, outperforming the unified approaches (i.e., VLMo and BEiT3) in terms of effectiveness and efficiency."
},
{
"venue": "ECIR",
"title": "Are Generative AI Agents Effective Personalized Financial Advisors?",
"authors": [
"Takehiro Takayanagi",
"Kiyoshi Izumi",
"Javier Sanz-Cruzado",
"Richard McCreadie",
"Iadh Ounis"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729897",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729897",
"abstract": "Large language model-based agents are becoming increasingly popular as a low-cost mechanism to provide personalized, conversational advice, and have demonstrated impressive capabilities in relatively simple scenarios, such as movie recommendations. But how do these agents perform in complex high-stakes domains, where domain expertise is essential and mistakes carry substantial risk? This paper investigates the effectiveness of LLM-advisors in the finance domain, focusing on three distinct challenges: (1) eliciting user preferences when users themselves may be unsure of their needs (2) providing personalized guidance for diverse investment preferences, and (3) leveraging advisor personality to build relationships and foster trust. Via a lab-based user study with 64 participants, we show that LLM-advisors often match human advisor performance when eliciting preferences, although they can struggle to resolve conflicting user needs. When providing personalized advice, the LLM was able to positively influence user behavior, but demonstrated clear failure modes. Our results show that accurate preference elicitation is key, otherwise, the LLM-advisor has little impact, or can even direct the investor toward unsuitable assets. More worryingly, users appear insensitive to the quality of advice being given, or worse these can have an inverse relationship. Indeed, users reported a preference for and increased satisfaction as well as emotional trust with LLMs adopting an extroverted persona, even though those agents provided worse advice."
},
{
"venue": "ECIR",
"title": "Benchmarking LLM-based Relevance Judgment Methods",
"authors": [
"Negar Arabzadeh",
"Charles L. A. Clarke"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730305",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730305",
"abstract": "Large Language Models (LLMs) are increasingly deployed in both academic and industry settings to automate the evaluation of information seeking systems, particularly by generating graded relevance judgments. Several studies report Kendall τ correlations exceeding 0.85 when comparing system rankings derived from human versus LLM-generated relevance labels. Previous work on LLM-based relevance assessment has primarily focused on replicating graded human relevance judgments through various prompting strategies. However, there has been limited exploration of alternative assessment methods or comprehensive comparative studies. In this paper, we systematically compare multiple LLM-based relevance assessment methods, including binary relevance judgments, graded relevance assessments, pairwise preference-based methods, and two nugget-based evaluation methods~-~document-agnostic and document-dependent. Wherever possible, we employ state-of-the-art tools and optimized prompts tailored for these methods. In addition to a traditional comparison based on system rankings using Kendall correlations, we also examine how well LLM judgments align with human preferences, as inferred from relevance grades. We conduct extensive experiments on datasets from three TREC Deep Learning tracks 2019, 2020 and 2021 as well as the ANTIQUE dataset, which focuses on non-factoid open-domain question answering. Beyond dataset-specific results, our work offers a practical methodology for evaluating diverse LLM-based relevance assessment methods. As part of our data release, we include relevance judgments generated by both an open-source (Llama3.2b) and a commercial (gpt-4o) model. Our goal is to reproduce various LLM-based relevance judgment methods to provide a comprehensive comparison. We release all the relevance judgments as a resource that establishes a baseline for future work, ensuring a level playing field for evaluation of LLM-based relevance judgments. All code, data, and resources are publicly available in our GitHub Repository at https://github.com/Narabzad/llm-relevance-judgement-comparison"
},
{
"venue": "ECIR",
"title": "Is Relevance Propagated from Retriever to Generator in RAG?",
"authors": [
"Feng Tian",
"Debasis Ganguly",
"Craig Macdonald"
],
"year": 2025,
"pdf_url": "https://eprints.gla.ac.uk/343843/3/343843.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/978-3-031-88708-6_3",
"abstract": ""
},
{
"venue": "ECIR",
"title": "DiSCo: LLM Knowledge Distillation for Efficient Sparse Retrieval in Conversational Search",
"authors": [
"Simon Lupart",
"Mohammad Aliannejadi",
"Evangelos Kanoulas"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729966",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729966",
"abstract": "Conversational Search (CS) involves retrieving relevant documents from a corpus while considering the conversational context, integrating retrieval with context modeling. Recent advancements in Large Language Models (LLMs) have significantly enhanced CS by enabling query rewriting based on conversational context. However, employing LLMs during inference poses efficiency challenges. Existing solutions mitigate this issue by distilling embeddings derived from human-rewritten queries, focusing primarily on learning the context modeling task. These methods, however, often separate the contrastive retrieval task from the distillation process, treating it as an independent loss term. To overcome these limitations, we introduce DiSCo (Distillation of Sparse Conversational retrieval), a novel approach that unifies retrieval and context modeling through a relaxed distillation objective. Instead of relying exclusively on representation learning, our method distills similarity scores between conversations and documents, providing more freedom in the representation space and better leveraging the contrastive nature of document relevance. Extensive experiments on Learned Sparse Retrieval (LSR) across five CS datasets demonstrate that DiSCo achieves substantial improvements in both in-domain and out-of-domain retrieval tasks, achieving up to a six-point gain in recall for out-of-domain datasets over state-of-the-art methods. Additionally, DiSCo employs a multi-teacher distillation strategy, using multiple LLMs as teachers, further enhancing performance and surpassing the individual teachers in in-domain settings. Furthermore, analysis of model sparsity reveals that DiSCo allows for more effective control over the sparsity of the trained models."
},
{
"venue": "ECIR",
"title": "Is Semantic Chunking Worth the Computational Cost?",
"authors": [
"Renyi Qu",
"Ruixuan Tu",
"Forrest Sheng Bao"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-naacl.114.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-naacl.114",
"abstract": "Recent advances in Retrieval-Augmented Generation (RAG) systems have popularized semantic chunking, which aims to improve retrieval performance by dividing documents into semantically coherent segments.Despite its growing adoption, the actual benefits over simpler fixed-size chunking, where documents are split into consecutive, fixed-size segments, remain unclear.This study systematically evaluates the effectiveness of semantic chunking using three common retrieval-related tasks: document retrieval, evidence retrieval, and retrievalbased answer generation.The results show that the computational costs associated with semantic chunking are not justified by consistent performance gains.These findings challenge the previous assumptions about semantic chunking and highlight the need for more efficient chunking strategies in RAG systems."
},
{
"venue": "ECIR",
"title": "Analyzing AI Evaluation Benchmarks Through Information Retrieval and Network Science",
"authors": [
"Soprano, Michael",
"Roitero, Kevin",
"Simeoni, Gaia",
"Lunardi, Riccardo",
"Mizzaro, Stefano"
],
"year": 2025,
"pdf_url": "https://doi.org/10.17605/osf.io/x2h6a",
"source": "openalex",
"doi": "https://doi.org/10.17605/osf.io/x2h6a",
"abstract": "Results and supplementary materials for the ECIR 2026 paper Analyzing AI Evaluation Benchmarks Through Information Retrieval and Network Science. The project supports a HITS-based bipartite graph analysis of LLM QA benchmarks, with a focus on benchmark bias, ranking stability, and leaderboard robustness."
},
{
"venue": "ECIR",
"title": "Preface:IR4U2 2024, the 1st Workshop on Information Retrieval for Understudied Users",
"authors": [
"Maria Soledad Pera",
"Federica Cena",
"Theo Huibers",
"Monica Landoni",
"Noemi Mauro",
"Emiliana Murgia"
],
"year": 2025,
"pdf_url": "https://research.utwente.nl/en/publications/ce54aaf1-4018-429f-b538-695951e79457",
"source": "openalex",
"doi": "",
"abstract": "In this manuscript, we provide a brief overview of the inaugural edition of the IR4U2 workshop at ECIR 2024. For this edition, we intentionally chose an in-person format, which not only allowed for the presentation of accepted contributions but also fostered highly interactive discussions and enabled networking among attendees to build a sense of community."
},
{
"venue": "ECIR",
"title": "SPRec: Self-Play to Debias LLM-based Recommendation",
"authors": [
"Chongming Gao",
"Ruijun Chen",
"Shuai Yuan",
"Kexin Huang",
"Yuanqing Yu",
"Xiangnan He"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714524",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714524",
"abstract": "Large language models (LLMs) have attracted significant attention in recommendation systems.Current work primarily applies supervised fine-tuning (SFT) to adapt the model for recommendation tasks.However, SFT on positive examples only limits the model's ability to align with user preference.To address this, researchers recently introduced Direct Preference Optimization (DPO), which explicitly aligns LLMs with user preferences using offline preference ranking data.However, we found that DPO inherently biases the model towards a few items, exacerbating the filter bubble issue and ultimately degrading user experience.In this paper, we propose SPRec, a novel self-play framework designed to mitigate over-recommendation and improve fairness without requiring additional data or manual intervention.In each self-play iteration, the model undergoes an SFT step followed by a DPO step, treating offline interaction data as positive samples and the predicted outputs from the previous iteration as negative samples.This effectively re-weights the DPO loss function using the model's logits, adaptively suppressing biased items.Extensive experiments on multiple real-world datasets demonstrate SPRec's effectiveness in enhancing recommendation accuracy and fairness.The code is available via https://github.com/RegionCh/SPRec."
},
{
"venue": "ECIR",
"title": "LoCal: Logical and Causal Fact-Checking with LLM-Based Multi-Agents",
"authors": [
"Jiatong Ma",
"Linmei Hu",
"Rang Li",
"Wenbo Fu"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714748",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714748",
"abstract": "With the development of social media, people are exposed to a vast amount of unverified information, making fact-checking particularly important. Existing fact-checking methods primarily encourage breaking down claims into more easily solvable sub-tasks, and deriving final answers through reasoning with external evidence. However, these models face logical issues regarding whether and how the sub-tasks can logically be combined to form the original claims, and encounter causal errors in the reasoning process due to insufficient evidence or hallucinations from LLMs. In addition, they often suffer from a lack of interpretability. In this paper, we propose Logical and Causal fact-checking (LoCal), a novel fact-checking framework based on multiple LLM-based agents. The usage of multi-agent systems is due to their increasingly demonstrated ability to perform complex tasks in a manner similar to humans. LoCal primarily consists of a decomposing agent, multiple reasoning agents, and two evaluating agents. Specifically, the decomposing agent first utilizes the in-context learning ability of LLMs to break down complex claims into simpler sub-tasks, including fact verification tasks and question answering tasks. Afterwards, two types of reasoning agents are respectively utilized to retrieve external knowledge to address the fact verification tasks that require comparative analysis skills, and the question answering tasks that necessitate the ability of information extraction from evidence. We then combine the sub-tasks and their corresponding responses to generate a solution for evaluation. In order to enhance logical and causal consistency, two evaluating agents are respectively employed to examine whether the generated solution is logically equivalent to the original claim and determine whether the solution still holds when challenged by the counterfactual label. The evaluating agents provide confidence degrees for the solutions based on the evaluation results and iteratively correct the logical and causal errors in the reasoning process. We evaluate LoCal on two challenging datasets, and the results show that LoCal significantly outperforms all the baseline models across different settings of evidence availability. In addition, LoCal offers better interpretability by providing a structured solution along with detailed evaluating processes. We believe LoCal will provide valuable insights for future misinformation detection."
},
{
"venue": "ECIR",
"title": "Thematic-LM: A LLM-based Multi-agent System for Large-scale Thematic Analysis",
"authors": [
"Tingrui Qiao",
"Caroline Walker",
"Chris Cunningham",
"Yun Sing Koh"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714595",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714595",
"abstract": "Thematic analysis (TA) is a widely used qualitative method for identifying underlying meanings within unstructured text. However, TA requires manual processes, which become increasingly labour-intensive and time-consuming as datasets grow. While large language models (LLMs) have been introduced to assist with TA on small-scale datasets, three key limitations hinder their effectiveness. First, current approaches often depend on interactions between an LLM agent and a human coder, a process that becomes challenging with larger datasets. Second, with feedback from the human coder, the LLM tends to mirror the human coder, which provides a narrower viewpoint of the data. Third, existing methods follow a sequential process, where codes are generated for individual samples without recalling previous codes and associated data, reducing the ability to analyse data holistically. To address these limitations, we propose Thematic-LM, an LLM-based multi-agent system for large-scale computational thematic analysis. Thematic-LM assigns specialised tasks to each agent, such as coding, aggregating codes, and maintaining and updating the codebook. We assign coder agents different identity perspectives to simulate the subjective nature of TA, fostering a more diverse interpretation of the data. We applied Thematic-LM to the Dreaddit dataset and the Reddit climate change dataset to analyse themes related to social media stress and online opinions on climate change. We evaluate the resulting themes based on trustworthiness principles in qualitative research. Our study reveals insights such as assigning different identities to coder agents promotes divergence in codes and themes."
},
{
"venue": "ECIR",
"title": "Constructing and Evaluating Declarative RAG Pipelines in PyTerrier",
"authors": [
"Craig Macdonald",
"Jinyuan Fang",
"Andrew Parry",
"Zaiqiao Meng"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730150",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730150",
"abstract": "Search engines often follow a pipeline architecture, where complex but effective reranking components are used to refine the results of an initial retrieval. Retrieval augmented generation (RAG) is an exciting application of the pipeline architecture, where the final component generates a coherent answer for the users from the retrieved documents. In this demo paper, we describe how such RAG pipelines can be formulated in the declarative PyTerrier architecture, and the advantages of doing so. Our PyTerrier-RAG extension for PyTerrier provides easy access to standard RAG datasets and evaluation measures, state-of-the-art LLM readers, and using PyTerrier's unique operator notation, easy-to-build pipelines. We demonstrate the succinctness of indexing and RAG pipelines on standard datasets (including Natural Questions) and how to build on the larger PyTerrier ecosystem with state-of-the-art sparse, learned-sparse, and dense retrievers, and other neural rankers."
},
{
"venue": "ECIR",
"title": "Natural language processing in the patent domain: a survey",
"authors": [
"Lekang Jiang",
"Stefan M. Goetz"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s10462-025-11168-z.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s10462-025-11168-z",
"abstract": "Abstract Patents, which encapsulate crucial technical and legal information in text form and referenced drawings, present a rich domain for natural language processing (NLP). As NLP technologies evolve, large language models (LLMs) have demonstrated outstanding capabilities in general text processing and generation tasks. However, the application of LLMs in the patent domain remains under-explored and under-developed due to the complexity of patents, particularly their language and legal framework. Understanding the unique characteristics of patent documents and related research in the patent domain becomes essential for researchers to apply these tools effectively. Therefore, this paper aims to equip NLP researchers with the essential knowledge to navigate this complex domain efficiently. We introduce the relevant fundamental aspects of patents to provide solid background information. In addition, we systematically break down the structural and linguistic characteristics unique to patents and map out how NLP can be leveraged for patent analysis and generation. Moreover, we demonstrate the spectrum of text-based and multimodal patent-related tasks, including nine patent analysis and four patent generation tasks."
},
{
"venue": "ECIR",
"title": "Towards Smarter Assessments: Enhancing Bloom’s Taxonomy Classification with a Bayesian-Optimized Ensemble Model Using Deep Learning and TF-IDF Features",
"authors": [
"Ali Alammary",
"S.A. Masoud"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2079-9292/14/12/2312/pdf?version=1749189080",
"source": "openalex",
"doi": "https://doi.org/10.3390/electronics14122312",
"abstract": "Bloom’s taxonomy provides a well-established framework for categorizing the cognitive complexity of assessment questions, ensuring alignment with course learning outcomes (CLOs). Achieving this alignment is essential for constructing meaningful and valid assessments that accurately measure student learning. However, in higher education, the large volume of questions that instructors must develop each semester makes manual classification of cognitive levels a time-consuming and error-prone process. Despite various attempts to automate this classification, the highest accuracy reported in existing research has not exceeded 93.5%, highlighting the need for further advancements in this area. Furthermore, the best-performing deep learning models only reached an accuracy of 86%. These results emphasize the need for improvement, particularly in the application of deep learning models, which have not been fully exploited for this task. In response to these challenges, our study explores a novel approach to enhance the accuracy of cognitive level classification. We leverage a combination of augmentation through synonym substitution, advanced feature extraction techniques utilizing DistilBERT and TF-IDF, and a robust ensemble model incorporating soft voting. These methods were selected to capture both semantic meaning and term frequency, allowing the model to benefit from contextual depth and statistical relevance. Additionally, Bayesian optimization is employed for hyperparameter tuning to refine the model’s performance further. The novelty of our approach lies in the fusion of sparse TF-IDF features with dense DistilBERT embeddings, optimized through Bayesian search across multiple classifiers. This hybrid design captures both term-level salience and deep contextual semantics, something not fully exploited in prior models focused solely on transformer architectures. Our soft-voting ensemble capitalizes on classifier diversity, yielding more stable and accurate results. Through this integrated approach outperformed previous configurations with an accuracy of 96%, surpassing the current state-of-the-art results and setting a new benchmark for automated cognitive level classification. These findings have significant implications for the development of high-quality, scalable assessments in educational settings."
},
{
"venue": "ECIR",
"title": "TIREx Tracker: The Information Retrieval Experiment Tracker",
"authors": [
"Thomas Hagen",
"Maik Fröbe",
"Jan Heinrich Reimer",
"Harrisen Scells",
"Matthias Hagen",
"Martin Potthast"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730297",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730297",
"abstract": "The reproducibility and transparency of retrieval experiments depends on the availability of information about the experimental setup. However, the manual collection of experiment metadata can be tedious, error-prone, and inconsistent, which calls for an automated systematic collection. Expanding ir_metadata, we present the TIREx tracker, a tool that records hardware configurations, power/CPU/RAM/GPU usage, and experiment/system versions. Implemented as a lightweight platform-independent C binary, the TIREx tracker integrates seamlessly into Python, Java, or C/C++ workflows and can be easily integrated into shard task submissions, as we demonstrate for the TIRA/TIREx platform. Code, binaries, and documentation of the TIREx tracker are publicly available at https://github.com/tira-io/tirex-tracker."
},
{
"venue": "ECIR",
"title": "Multi-Domain Controversial Text Detection Based on a Machine Learning and Deep Learning Stacked Ensemble",
"authors": [
"Jiadi Liu",
"Zhuodong Liu",
"Qiaoqi Li",
"Weihao Kong",
"Xiangyu Li"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2227-7390/13/9/1529/pdf?version=1746539601",
"source": "openalex",
"doi": "https://doi.org/10.3390/math13091529",
"abstract": "Due to the rapid proliferation of social media and online reviews, the accurate identification and classification of controversial texts has emerged as a significant challenge in the field of natural language processing. However, traditional text-classification methodologies frequently encounter critical limitations, such as feature sensitivity and inadequate generalization capabilities. This results in a notably suboptimal performance when confronted with diverse controversial content. To address these substantial limitations, this paper proposes a novel controversial text-detection framework based on stacked ensemble learning to enhance the accuracy and robustness of text classification. Firstly, considering the multidimensional complexity of textual features, we integrate comprehensive feature engineering, i.e., encompassing word frequency, statistical metrics, sentiment analysis, and comment tree structure features, as well as advanced feature selection methodologies, particularly lassonet, i.e., a neural network with feature sparsity, to effectively address dimensionality challenges while enhancing model interpretability and computational efficiency. Secondly, we design a two-tier stacked ensemble architecture, which not only combines the strengths of multiple machine learning algorithms, e.g., gradient-boosted decision tree (GBDT), random forest (RF), and extreme gradient boosting (XGBoost), with deep learning models, e.g., gated recurrent unit (GRU) and long short-term memory (LSTM), but also implements the support vector machine (SVM) for efficient meta-learning. Furthermore, we systematically compare three hyperparameter optimization algorithms, including the sparrow search algorithm (SSA), particle swarm optimization (PSO), and Bayesian optimization (BO). The experimental results demonstrate that the SSA exhibits a superior performance in exploring high-dimensional parameter spaces. Extensive experimentation across diverse topics and domains also confirms that our proposed methodology significantly outperforms the state-of-the-art approaches."
},
{
"venue": "ECIR",
"title": "Re2LLM: Reflective Reinforcement Large Language Model for Session-based Recommendation",
"authors": [
"Ziyan Wang",
"Yingpeng Du",
"Zhu Sun",
"Haoyan Chua",
"Kaidong Feng",
"Wenya Wang",
"Jie Zhang"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/33399/35554",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i12.33399",
"abstract": "Emerging advancements in large language models (LLMs) show significant potential for enhancing recommendations. However, prompt-based methods often struggle to find ideal prompts without task-specific feedback, while fine-tuning-based methods are hindered by high computational demands and dependence on open-source backbones. To address these challenges, we propose a Reflective Reinforcement Large Language Model (Re2LLM) for session-based recommendation, which refines LLMs to generate and utilize specialized knowledge effectively and efficiently. Specifically, we first devise the Reflective Exploration Module to extract and present knowledge in a form that LLMs can easily process. This module enables LLMs to reflect on their recommendation mistakes and construct a hint knowledge base to rectify them effectively. Next, we design the Reinforcement Utilization Module to train a lightweight retrieval agent that elicits correct LLM reasoning. This module recognizes hints as signals to facilitate LLM recommendations and learns to select appropriate hints from the constructed knowledge base using task-specific feedback efficiently. Lastly, we conduct experiments on real-world datasets and demonstrate the superiority of our Re2LLM over state-of-the-art methods."
},
{
"venue": "ECIR",
"title": "Evolving techniques in sentiment analysis: a comprehensive review",
"authors": [
"M. R. Pavan Kumar",
"Lal Khan",
"Hsien-Tsung Chang"
],
"year": 2025,
"pdf_url": "https://peerj.com/articles/cs-2592.pdf",
"source": "openalex",
"doi": "https://doi.org/10.7717/peerj-cs.2592",
"abstract": "With the rapid expansion of social media and e-commerce platforms, an unprecedented volume of user-generated content has emerged, offering organizations, governments, and researchers invaluable insights into public sentiment. Yet, the vast and unstructured nature of this data challenges traditional analysis methods. Sentiment analysis, a specialized field within natural language processing, has evolved to meet these challenges by automating the detection and categorization of opinions and emotions in text. This review comprehensively examines the evolving techniques in sentiment analysis, detailing foundational processes such as data gathering and feature extraction. It explores a spectrum of methodologies, from classical word embedding techniques and machine learning algorithms to recent contextual embedding and advanced transformer models like Generative Pre-trained Transformer (GPT), Bidirectional Encoder Representations from Transformers (BERT), and T5. With a critical comparison of these methods, this article highlights their appropriate uses and limitations. Additionally, the review provides a thorough overview of current trends, insights into future directions, and a critical exploration of unresolved challenges. By synthesizing these developments, this review equips researchers with a solid foundation for assessing the current state of sentiment analysis and guiding future advancements in this dynamic field."
},
{
"venue": "ECIR",
"title": "Dynamic Superblock Pruning for Fast Learned Sparse Retrieval",
"authors": [
"Parker Carlson",
"Wentai Xie",
"Shanxiu He",
"Tao Yang"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730183",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730183",
"abstract": "This paper proposes superblock pruning (SP) during top-k online document retrieval for learned sparse representations. SP structures the sparse index as a set of superblocks on a sequence of document blocks and conducts a superblock-level selection to decide if some superblocks can be pruned before visiting their child blocks. SP generalizes the previous flat block or cluster-based pruning, allowing the early detection of groups of documents that cannot or are less likely to appear in the final top-k list. SP can accelerate sparse retrieval in a rank-safe or approximate manner under a high-relevance competitiveness constraint. Our experiments show that the proposed scheme significantly outperforms state-of-the-art baselines on MS MARCO passages on a single-threaded CPU."
},
{
"venue": "ECIR",
"title": "Human-Centered and Sustainable Recommender Systems",
"authors": [
"Allegra De Filippo",
"Ludovico Boratto",
"Giuseppe Spillo"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3708319.3727553",
"source": "openalex",
"doi": "https://doi.org/10.1145/3708319.3727553",
"abstract": "This tutorial explores the intersection of sustainability and recommender systems, focusing on aligning user needs and values with sustainable practices.It emphasizes two dimensions: (1) understanding and modeling users to deliver more sustainable recommendations; and (2) fostering sustainability through system design and functionality.Participants will learn how recommender systems can encourage sustainable behaviors and how to enhance system efficiency while minimizing resource consumption and ethical challenges.Through theoretical insights and hands-on sessions, this tutorial proposes discussion and actionable strategies to design human-centered, sustainable recommender systems, addressing both societal impact and technological responsibility."
},
{
"venue": "ECIR",
"title": "On Assessing the Performance of LLMs for Target-Level Sentiment Analysis in Financial News Headlines",
"authors": [
"Iftikhar Muhammad",
"Marco Rospocher"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/1999-4893/18/1/46/pdf?version=1736780560",
"source": "openalex",
"doi": "https://doi.org/10.3390/a18010046",
"abstract": "The importance of sentiment analysis in the rapidly evolving financial markets is widely recognized for its ability to interpret market trends and inform investment decisions. This study delves into the target-level financial sentiment analysis (TLFSA) of news headlines related to stock. The study compares the performance in the TLFSA task of various sentiment analysis techniques, including rule-based models (VADER), fine-tuned transformer-based models (DistilFinRoBERTa and Deberta-v3-base-absa-v1.1) as well as zero-shot large language models (ChatGPT and Gemini). The dataset utilized for this analysis, a novel contribution of this research, comprises 1476 manually annotated Bloomberg headlines and is made publicly available (due to copyright restrictions, only the URLs of Bloomberg headlines with the manual annotations are provided; however, these URLs can be used with a Bloomberg terminal to reconstruct the complete dataset) to encourage future research on this subject. The results indicate that the fine-tuned Deberta-v3-base-absa-v1.1 model performs better across all evaluation metrics than other evaluated models in TLFSA. However, LLMs such as ChatGPT-4, ChatGPT-4o, and Gemini 1.5 Pro provide similar performance levels without the need for task-specific fine-tuning or additional training. The study contributes to assessing the performance of LLMs for financial sentiment analysis, providing useful insights into their possible application in the financial domain."
},
{
"venue": "ECIR",
"title": "Evaluating Multimedia Technologies An EDAS-Based Analysis of Quality, Accessibility, Cost, and Energy Efficiency",
"authors": [
"Chandrasekar Raja",
"Devipriya Mani",
"M Ramachandran",
"Nathiya Murali",
"Robinson Joel",
"M",
"G Manikandan",
"G Bhuvaneswari",
"P Shanthakumar",
"Lina Markauskaite",
"Nayeemuddin",
"Anitha Anandhan",
"Liyana Shuib",
"Maizatul Akmar Ismail",
"Ghulam Mujtaba",
"Chunho Lee",
"Miodrag Potkonjak",
"William Mangione-Smith",
"D Annapurna",
"K Raja",
"K Venugopal",
"L Patnaik",
"Neha Bharani",
"Adel Elmaghraby",
"M Mehmed",
"Mark Kantardzic",
"Wachowiak",
"Wayne Zhao",
"Jing Xin",
"Jianshu Jiang",
"Jing Weng",
"Ee-Peng He",
"Hongfei Lim",
"Xiaoming Yan",
"Li",
"Konduru Madhavi",
"Kommanaboyina Reddy",
"Lakshmi Sundara",
"Rambabu Soujanya",
"Inaganti",
"Ayyappa Kondru",
"Sreekanth Swamy",
"Madhu Yalavarthi",
"Munagala",
"Muthu Balaanand",
"N Karthikeyan",
"S Karthik",
"Kriti Bhushan",
"Brij Gupta",
"Zheng Xu",
"Yunhuai Liu",
"Hui Zhang",
"Xiangfeng Luo",
"Lin Mei",
"Chuanping Hu",
"Jun Li",
"Zhi He",
"Javier Plaza",
"Shutao Li",
"Jinfen Chen",
"Henglin Wu",
"Yandong Wang",
"Yu Liu",
"Neha Bharani",
"Abhay Kothari",
"Chih- Tsai",
"Mao-Yuan Fong",
"Chen",
"Han Hu",
"Yonggang Wen",
"Tat-Seng Chua",
"Xuelong Li",
"Mostafa Ezzat",
"Mohamed Ahmed",
"Sultan Abd El Ghany",
"Almotairi",
"A-M Mohammed",
"Salem",
"Muhammad Alam",
"Fazeel Abid",
"Cong Guangpei",
"L Yunrong",
"Rambabu Inaganti",
"Sreekanth Yalavarthi",
"S Suhasini",
"J Sheelalavanya",
"M Parameswari",
"G Manikandan",
"S Nissi",
"P Jyothi",
"G Pradeepini",
"P Varma",
"Prathap Siva",
"Yamini Adimoolam",
"Anantharamaiah Lahari Marna",
"Vengala",
"Vs Divya",
"Sundar",
"Pavan Ram",
"Kumar"
],
"year": 2025,
"pdf_url": "https://doi.org/10.46632/jdaai/4/3/3",
"source": "openalex",
"doi": "https://doi.org/10.46632/jdaai/4/3/3",
"abstract": "Multimedia technology has transformed the way information is presented and consumed, revolutionizing communication, entertainment, education, and various industries. This paper explores the evolution of multimedia technology, its components, and the advancements that have shaped its current state. It delves into various aspects of multimedia, including graphics, audio, video, and interactive elements, discussing their role in creating immersive and engaging experiences. The paper also highlights the impact of multimedia on society and business, examining its applications in fields such as advertising, education, virtual reality, and gaming. Furthermore, it addresses the challenges and opportunities posed by the rapid evolution of multimedia technology, including issues related to content creation, distribution, and copyright. By analyzing the trajectory of multimedia technology, this paper offers insights into its potential future developments and the implications they might have on our digital landscape. Multimedia technology has revolutionized communication by enabling the creation of engaging and interactive content. Research in this field helps uncover innovative ways to convey information effectively, fostering better communication in diverse contexts, from advertising and journalism to social media and online collaboration. The integration of multimedia elements in education enhances the learning experience. Research in this area explores how multimedia can be used to create dynamic and immersive educational content, catering to different learning styles and making complex concepts more understandable. This research enables us to fully grasp the potential of multimedia and leverage it for positive societal, economic, and technological advancements. The EDAS score primarily based on the space from the suggest agreement machine is the installed energy for a manufacturing plant. Experts' critiques and derived numbers do not trust each different concerning solar energy and geothermal electricity. Alternative taken as Technology A, Technology B, Technology C, Technology D, Technology Evaluation parameters taken as improved quality, Increased Accessibility, cost, Energy consumption kwh. Technology A is ranked at first position and Technology D is ranked at fifth position"
},
{
"venue": "ECIR",
"title": "Efficient Constant-Space Multi-vector Retrieval",
"authors": [
"Sean MacAvaney",
"Antonio Mallia",
"Nicola Tonellotto"
],
"year": 2025,
"pdf_url": "https://eprints.gla.ac.uk/view/author/60888.html>",
"source": "openalex",
"doi": "https://doi.org/10.1007/978-3-031-88714-7_22",
"abstract": ""
},
{
"venue": "ECIR",
"title": "Lost in Transliteration: Bridging the Script Gap in Neural IR",
"authors": [
"Andreas Chari",
"Iadh Ounis",
"Sean MacAvaney"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730226",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730226",
"abstract": "Most human languages use scripts other than the Latin alphabet. Search users in these languages often formulate their information needs in a transliterated -usually Latinized- form for ease of typing. For example, Greek speakers might use Greeklish, and Arabic speakers might use Arabizi. This paper shows that current search systems, including those that use multilingual dense embeddings such as BGE-M3, do not generalise to this setting, and their performance rapidly deteriorates when exposed to transliterated queries. This creates a ''script gap'' between the performance of the same queries when written in their native or transliterated form. We explore whether adapting the popular ''translate-train'' paradigm to transliterations can enhance the robustness of multilingual Information Retrieval (IR) methods and bridge the gap between native and transliterated scripts. https://github.com/andreaschari/transliterations By exploring various combinations of non-Latin and Latinized query text for training, we investigate whether we can enhance the capacity of existing neural retrieval techniques and enable them to apply to this important setting. We show that by further fine-tuning IR models on an even mixture of native and Latinized text, they can perform this cross-script matching at nearly the same performance as when the query was formulated in the native script. Out-of-domain evaluation and further qualitative analysis show that transliterations can also cause queries to lose some of their nuances, motivating further research."
},
{
"venue": "ECIR",
"title": "MULTIDIMENSIONAL ECONOMIC COMPLEXITY AND ECONOMIC GROWTH: EVIDENCE FROM THE 10 MOST COMPLEX COUNTRIES",
"authors": [
"Seyhun Tutgun"
],
"year": 2025,
"pdf_url": "https://dergipark.org.tr/en/download/article-file/4419303",
"source": "openalex",
"doi": "https://doi.org/10.17130/ijmeb.1596858",
"abstract": "Economic complexity, reflecting the collective know-how and the depth of national capability sets, features prominently as a key concept within the economic growth literature. This study addresses the multidimensional nature of economic complexity by analyzing the interplay between economic growth (GDP per capita), the Economic Research Complexity Index (ECIR) reflecting knowledge generation capacity, the Economic Trade Complexity Index (ECITRA) capturing the sophistication of a nation's export structure, and the Economic Technology Complexity Index (ECITEC) indicating technological advancement. To provide a more robust analysis, population (POP) and human capital (HUM) are included as control variables. The study investigates these relationships through three distinct models, each focusing on one dimension of complexity. The analysis focuses specifically on the world's 10 most complex economies: Austria, Czechia, Germany, Japan, Singapore, Slovenia, South Africa, Sweden, Switzerland, and the United Kingdom. The relationship between economic growth and multidimensional economic complexity is analyzed using annual data for the period 1999-2021. Second-generation panel data analysis techniques are used in the study, where for each of the three models, cross-section dependence, homogeneity/heterogeneity, unit root, and cointegration tests are applied respectively. After finding that there is a cointegration relationship between the series, the long-run coefficients are estimated using the Common Correlated Effects (CCE) and Augmented Mean Group (AMG) estimators, which are robust to cross-sectional dependence and heterogeneity. Furthermore, the causal relationships between the variables are examined using the Dumitrescu-Hurlin panel causality test. The empirical findings indicate significant heterogeneity in the coefficients of ECIR, ECITRA, and ECITEC, both at the panel-average level and, more strikingly, for individual countries. Ultimately, the findings reinforce the importance, relevant for all nations, of cultivating complex capabilities and aligning human capital with the demands of the knowledge economy as crucial elements for achieving long-term development."
},
{
"venue": "ECIR",
"title": "Guiding Retrieval Using LLM-Based Listwise Rankers",
"authors": [
"Mandeep Rathee",
"Sean MacAvaney",
"Avishek Anand"
],
"year": 2025,
"pdf_url": "https://eprints.gla.ac.uk/view/author/60888.html>",
"source": "openalex",
"doi": "https://doi.org/10.1007/978-3-031-88708-6_15",
"abstract": ""
},
{
"venue": "ECIR",
"title": "An intelligent guided troubleshooting method for aircraft based on HybirdRAG",
"authors": [
"Xiaoyue Xie",
"Xilang Tang",
"Simin Gu",
"Lijie Cui"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-02643-2.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-02643-2",
"abstract": "To enhance aircraft fault diagnosis efficiency, this paper proposes HybridRAG, an intelligent-guided troubleshooting framework that integrates knowledge graphs and large language models (LLMs). Unlike conventional retrieval-augmented generation (RAG) methods that rely on single-modal retrieval, HybridRAG adopts a multi-dimensional retrieval strategy, combining graph-based reasoning with both vector-based and BM25-based text retrieval techniques. This hybrid approach ensures comprehensive extraction of relevant information from both unstructured text and structured fault graphs, enhancing diagnostic precision, relevance, and robustness. Experimental results demonstrate that HybridRAG achieves an F1 score improvement of at least 4% and reduces hallucination rates by over 7% compared to mainstream RAG baselines. These advancements, combined with its unique integration of multi-modal retrieval, position HybridRAG as a novel framework for addressing complex aircraft maintenance challenges. Additionally, the paper presents an agent-based intelligent troubleshooting assistant that supports more interactive, adaptive, and flexible diagnostic Q&A, providing maintenance personnel with a significant advanced intelligent, context-aware diagnostic tool."
},
{
"venue": "ECIR",
"title": "CRS Arena: Crowdsourced Benchmarking of Conversational Recommender Systems",
"authors": [
"Nolwenn Bernard",
"Hideaki Joko",
"Faegheh Hasibi",
"Krisztian Balog"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2412.10514",
"source": "openalex",
"doi": "https://doi.org/10.1145/3701551.3704120",
"abstract": "We introduce CRS Arena, a research platform for scalable benchmarking of Conversational Recommender Systems (CRS) based on human feedback. The platform displays pairwise battles between anonymous conversational recommender systems, where users interact with the systems one after the other before declaring either a winner or a draw. CRS Arena collects conversations and user feedback, providing a foundation for reliable evaluation and ranking of CRSs. We conduct experiments with CRS Arena on both open and closed crowdsourcing platforms, confirming that both setups produce highly correlated rankings of CRSs and conversations with similar characteristics. We release CRSArena-Dial, a dataset of 474 conversations and their corresponding user feedback, along with a preliminary ranking of the systems based on the Elo rating system. The platform is accessible at https://iai-group-crsarena.hf.space/."
},
{
"venue": "ECIR",
"title": "The Great Nugget Recall: Automating Fact Extraction and RAG Evaluation with Large Language Models",
"authors": [
"Ronak Pradeep",
"Nandan Thakur",
"Shivani Upadhyay",
"Daniel Campos",
"Nick Craswell",
"Ian Soboroff",
"Hoa Trang Dang",
"Jimmy Lin"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730090",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730090",
"abstract": "Large Language Models (LLMs) have significantly enhanced the capabilities of information access systems, especially with retrieval-augmented generation (RAG). Nevertheless, the evaluation of RAG systems remains a barrier to continued progress, a challenge we tackle in this work by proposing an automatic evaluation framework that is validated against human annotations. We believe that the nugget evaluation methodology provides a solid foundation for evaluating RAG systems. This approach, originally developed for the TREC Question Answering (QA) Track in 2003, evaluates systems based on atomic facts that should be present in good answers. Our efforts focus on ''refactoring'' this methodology, where we describe the AutoNuggetizer framework that specifically applies LLMs to both automatically create nuggets and automatically assign nuggets to system answers. In the context of the TREC 2024 RAG Track, we calibrate a fully automatic approach against strategies where nuggets are created manually or semi-manually by human assessors and then assigned manually to system answers. Based on results from a community-wide evaluation, we observe strong agreement at the run level between scores derived from fully automatic nugget evaluation and human-based variants. The agreement is stronger when individual framework components such as nugget assignment are automated independently. This suggests that our evaluation framework provides tradeoffs between effort and quality that can be used to guide the development of future RAG systems. However, further research is necessary to refine our approach, particularly in establishing robust per-topic agreement to diagnose system failures effectively."
},
{
"venue": "ECIR",
"title": "Artifact Sharing for Information Retrieval Research",
"authors": [
"Sean MacAvaney"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730147",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730147",
"abstract": "Sharing artifacts-such as trained models, pre-built indexes, and the code to use them-aids in reproducibility efforts by allowing researchers to validate intermediate steps and improves the sustainability of research by allowing multiple groups to build off one another's prior computational work. Although there are de facto consensuses on how to share research code (through a git repository linked to from publications) and trained models (via HuggingFace Hub), there is no consensus for other types of artifacts, such as built indexes. Given the practical utility of using shared indexes, researchers have resorted to self-hosting these resources or performing ad hoc file transfers upon request, ultimately limiting the artifacts' discoverability and reuse. This demonstration introduces a flexible and interoperable way to share artifacts for Information Retrieval research, improving both their accessibility and usability."
},
{
"venue": "ECIR",
"title": "Treatment of Urolithiasis: A Comprehensive Review",
"authors": [
"Živka Dika",
"Marjan Marić",
"Marijana Živko"
],
"year": 2025,
"pdf_url": "https://www.emjreviews.com/wp-content/uploads/2025/05/Treatment-of-Urolithiasis-A-Comprehensive-Review.pdf",
"source": "openalex",
"doi": "https://doi.org/10.33590/emjurol/nbza7146",
"abstract": "Urolithiasis is a highly prevalent and multifactorial disease, representing the most common urological condition globally. Its incidence continues to rise, posing a significant public health challenge. Management strategies are tailored to individual patients, considering factors such as stone size, composition, location, and underlying metabolic conditions. This review provides a comprehensive overview of both pharmacological and surgical approaches to urolithiasis, emphasising recent advancements and emerging technologies. Surgical treatments, particularly minimally invasive procedures, have shown considerable improvement, offering highly effective solutions with reduced morbidity. While surgical interventions are essential, pharmacological therapies play a key role in preventing recurrence and addressing metabolic abnormalities that contribute to stone formation. Advances in understanding the molecular mechanisms and pathophysiology of urolithiasis are crucial for developing targeted therapies. A holistic approach that integrates advanced surgical techniques with pharmacological interventions, tailored to individual metabolic profiles, is essential for optimising patient outcomes. Personalised management, supported by regular monitoring of urinary pH, metabolic profiles, and adherence to treatment regimens, is vital to reducing recurrence and improving the quality of life for patients with urolithiasis."
},
{
"venue": "ECIR",
"title": "Longer‐term effects of intraoperative cone‐beam computed tomography in percutaneous nephrolithotomy: 18‐month retrospective randomised controlled trial analysis",
"authors": [
"Chris A. Suijker",
"Riemer A. Kingma",
"R. van Ee",
"Emelie N. Steffens",
"Emanuela Altobelli",
"Mieke T. J. Bus",
"Igle J. de Jong",
"Stijn Roemeling"
],
"year": 2025,
"pdf_url": "https://onlinelibrary.wiley.com/doi/pdfdirect/10.1111/bju.16859",
"source": "openalex",
"doi": "https://doi.org/10.1111/bju.16859",
"abstract": "OBJECTIVES: To assess the longer-term impact of intraoperative cone-beam computed tomography (CBCT) on stone-related morbidity after percutaneous nephrolithotomy (PCNL), since intraoperative CBCT allows for the detection and removal of residual fragments during the same procedure, improving stone clearance and thereby potentially diminishing stone-related morbidity. PATIENTS AND METHODS: This study was a retrospective analysis of a previously conducted single-centre randomised controlled trial at a tertiary complex endourology centre, in which patients were randomised intraoperatively to PCNL with intraoperative CBCT or conventional PCNL. We analysed 18-month follow-up data to assess differences in stone-related events (SREs), including re-interventions, emergency department visits, and hospital admissions. Stone-free rates and time to stone recurrence, as determined by follow-up CT scans, were also evaluated. RESULTS: The CBCT group (n = 80) had a significantly lower detection rate of new or residual fragments >4 mm (hazard ratio [HR] 0.61, 95% confidence interval [CI] 0.38-0.97), with 29 (36%) cases during 18 months of follow-up compared to 40 (50%) cases in the conventional PCNL group (n = 80). The restricted mean (standard deviation [SD]) time to fragment detection was 420 (44) days in the CBCT group vs 318 (53) days in the conventional PCNL group. In the CBCT group, 15 (19%) cases experienced 26 SREs, compared to 23 (29%) cases with 39 SREs in the conventional PCNL group. The restricted mean (SD) time to SRE was 499 (26) days for CBCT cases, compared to 447 (39) days for conventional PCNL cases. The rate of SREs did not decrease significantly when comparing CBCT-PCNL to conventional PCNL (HR 0.61, 95% CI 0.32-1.17). CONCLUSION: This study found 10% fewer patients with SREs in the 18-month period after PCNL with intraoperative CBCT compared to conventional PCNL. This difference is likely due to the notable increase in stone-free rates following a single PCNL with intraoperative CBCT."
},
{
"venue": "ECIR",
"title": "Predicting RAG Performance for Text Completion",
"authors": [
"Oz Huly",
"David Carmel",
"Oren Kurland"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730062",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730062",
"abstract": "We address the challenge of predicting the performance of using retrieval augmented generation (RAG) in large language models (LLMs) for the task of text completion; specifically, we predict the perplexity gain attained by applying RAG. We present novel supervised post-retrieval prediction methods that utilize the specific characteristics of the text completion setting. Our predictors substantially outperform a wide variety of prediction methods originally proposed for ad hoc document retrieval. We then show that integrating our post-retrieval predictors with recently proposed post-generation predictors - i.e., those analyzing the next-token distribution - is of much merit: the resultant prediction quality is statistically significantly better than that of using the post-generation predictors alone. Finally, we show that our post-retrieval predictors are as effective as post-generation predictors for selective application of RAG. This finding is of utmost importance in terms of efficiency of selective RAG."
},
{
"venue": "ECIR",
"title": "Large Language Models for Information Retrieval: Challenges and Chances",
"authors": [
"Timo Breuer",
"Sameh Frihat",
"Norbert Fuhr",
"Dirk Lewandowski",
"Philipp Schaer",
"Ralf Schenkel"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s13222-025-00503-x.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s13222-025-00503-x",
"abstract": "Abstract The rapid advancement of Large Language Models (LLMs) has introduced a paradigm shift in Information Retrieval (IR), moving beyond conventional keyword queries and ranked result lists. LLMs now play a critical role in the evolution of IR technologies and introduce new interaction forms like Retrieval-Augmented Generation, which is a more dynamic and interactive retrieval process that integrates various aspects of Information Access, like Question Answering, into the dialog between a searcher and the search engine. We explore the multi-faceted impact of LLMs on IR, particularly in three distinct layers where they have become an integral part of the retrieval process, namely the retrieval system and processing pipeline that can make use of a richer semantic representation using advanced language models, the interaction layer, and the broader IR ecosystem. For the latter, we focus on evaluation issues as well as bias, fairness, and ethical concerns. We also highlight some recent cases of using LLMs in the medical domain to demonstrate the impact on one specific domain."
},
{
"venue": "ECIR",
"title": "Uncovering Cross-Domain Recommendation Ability of Large Language Models",
"authors": [
"Xinyi Liu",
"Ruijie Wang",
"Dachun Sun",
"Dilek Hakkani Tür",
"Tarek Abdelzaher"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3701716.3717850",
"source": "openalex",
"doi": "https://doi.org/10.1145/3701716.3717850",
"abstract": "Cross-Domain Recommendation (CDR) seeks to enhance item retrieval in low-resource domains by transferring knowledge from high-resource domains. While recent advancements in Large Language Models (LLMs) have demonstrated their potential in Recommender Systems (RS), their ability to effectively transfer domain knowledge for improved recommendations remains underexplored. To bridge this gap, we propose LLM4CDR, a novel CDR pipeline that constructs context-aware prompts by leveraging users' purchase history sequences from a source domain along with shared features between source and target domains. Through extensive experiments, we show that LLM4CDR achieves strong performance, particularly when using LLMs with large parameter sizes and when the source and target domains exhibit smaller domain gaps. For instance, incorporating CD & Vinyl purchase history for recommendations in Movies & TV yields a 64.28% MAP@1 improvement. We further investigate how key factors-source domain data, domain gap, prompt design, and LLM size-impact LLM4CDR's effectiveness in CDR tasks. Our results highlight that LLM4CDR excels when leveraging a single, closely related source domain and benefits significantly from larger LLMs. These insights pave the way for future research on LLM-driven cross-domain recommendations."
},
{
"venue": "ECIR",
"title": "Percutaneous Nephrolithotomy",
"authors": [
"Antun Gršković",
"Iva Bukša",
"Aleksandar Fišer"
],
"year": 2025,
"pdf_url": "https://hrcak.srce.hr/file/480479",
"source": "openalex",
"doi": "https://doi.org/10.21860/medflum2025_332185",
"abstract": "One of the most complex procedures in urology is percutaneous nephrolithotomy. This procedure can be performed in the patient’s prone as well as supine position. The position of the patient during percutaneous nephrolithotomy affects the safety of the procedure, anaesthetic risk, hospitalization time, surgery time, and postoperative complications. The advantages of the supine PCNL are easier patient positioning, shorter surgery time, the possibility of spinal anaesthesia, lower anaesthetic risk in obese patients and patients with high cardiovascular risk, lower risk of injuries to the nervous and musculoskeletal system, and lower risk of infections and sepsis. In the supine PCNL, it is possible to perform simultaneous bilateral endoscopic surgery (SBES) and endoscopic combined intrarenal surgery (ECIRS) as well. The advantages of the prone PCNL are shorter puncture channel, which results in a lower risk of injury to the adjacent organs, the possibility of upper renal calyx puncture, and greater mobility when handling the nephroscope. Although the prone position is used more often today, modern knowledge and clinical experience should encourage us to leave our comfort zone and, considering the patient’s characteristics, critically evaluate whether we will perform the procedure in the supine or prone position."
},
{
"venue": "ECIR",
"title": "Fair Exposure Allocation Using Generative Query Expansion",
"authors": [
"Thomas Jaenich",
"Graham McDonald",
"Iadh Ounis"
],
"year": 2025,
"pdf_url": "https://eprints.gla.ac.uk/344597/2/344597.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/978-3-031-88717-8_20",
"abstract": ""
},
{
"venue": "ECIR",
"title": "Semi-automated annotation for video-based beef cattle behavior recognition",
"authors": [
"Zhiyong Cao",
"Chen Li",
"Xiujuan Yang",
"Shuai Zhang",
"Luo Ling",
"Hao Wang",
"Hongbo Zhao"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-01948-6.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-01948-6",
"abstract": "In this paper, a video-based behavioural recognition dataset for beef cattle is constructed. The dataset covers five behaviours of beef cattle: standing, lying, drinking, feeding, and ruminating. Six beef cows in a captive barn were selected and monitored for 168 h. Different light conditions and nighttime data were considered. The dataset was collected by one surveillance video camera. The data collection process required deploying cameras, memory, routers and laptops. Data annotation was automated using the YOLOv8 target detection model and the ByteTrack multi-target tracking algorithm to annotate each beef cow's coordinates and identity codes. The FFmpeg tool cut out individual beef cow video clips and manually annotated them with behavioural labels. The dataset includes 500 video clips, 2000 image recognition samples, over 4000 target tracking samples, and over 10G of frame sequence images. 4974 video data of different behavioural types are labelled, totalling about 14 h. Based on this, a TimeSformer multi-behaviour recognition model for beef cattle based on video understanding is proposed as a baseline evaluation model. The experimental results show that the model can effectively learn the corresponding category labels from the behavioural category data of the dataset, with an average recognition accuracy of 90.33% on the test set. In addition, a data enhancement and oversampling strategy was adopted to solve the data imbalance problem and reduce the risk of model overfitting. The dataset provides a data basis for studying beef cattle behaviour recognition. It is of great significance for the intelligent perception of beef cattle health status and improvement of farming efficiency."
},
{
"venue": "ECIR",
"title": "Retrieval And Structuring Augmented Generation with Large Language Models",
"authors": [
"Pengcheng Jiang",
"Siru Ouyang",
"Yizhu Jiao",
"Ming Zhong",
"Runchu Tian",
"Jiawei Han"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3711896.3736557",
"source": "openalex",
"doi": "https://doi.org/10.1145/3711896.3736557",
"abstract": "Large Language Models (LLMs) have revolutionized natural language processing with their remarkable capabilities in text generation and reasoning. However, these models face critical challenges when deployed in real-world applications, including hallucination generation, outdated knowledge, and limited domain expertise. Retrieval And Structuring (RAS) Augmented Generation addresses these limitations by integrating dynamic information retrieval with structured knowledge representations. This survey (1) examines retrieval mechanisms including sparse, dense, and hybrid approaches for accessing external knowledge; (2) explore text structuring techniques such as taxonomy construction, hierarchical classification, and information extraction that transform unstructured text into organized representations; and (3) investigate how these structured representations integrate with LLMs through prompt-based methods, reasoning frameworks, and knowledge embedding techniques. It also identifies technical challenges in retrieval efficiency, structure quality, and knowledge integration, while highlighting research opportunities in multimodal retrieval, cross-lingual structures, and interactive systems. This comprehensive overview provides researchers and practitioners with insights into RAS methods, applications, and future directions."
},
{
"venue": "ECIR",
"title": "CLERC: A Dataset for U. S. Legal Case Retrieval and Retrieval-Augmented Analysis Generation",
"authors": [
"Abe Hou",
"Orion Weller",
"Guanghui Qin",
"Eugene Yang",
"Dawn Lawrie",
"Nils Holzenberger",
"Andrew Blair-Stanek",
"Benjamin Van Durme"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-naacl.441.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-naacl.441",
"abstract": "Abe Bohan Hou, Orion Weller, Guanghui Qin, Eugene Yang, Dawn Lawrie, Nils Holzenberger, Andrew Blair-Stanek, Benjamin Van Durme. Findings of the Association for Computational Linguistics: NAACL 2025. 2025."
},
{
"venue": "ECIR",
"title": "Tearing down inequalities in the healthcare system across Europe: the BEACON project",
"authors": [
"Chrysanthi Koukoutzeli",
"Giulia Ferraris",
"Veronica Coppini",
"Maria Vittoria Ferrari",
"Elisa Fragale",
"Dario Trapani",
"Ida Minchella",
"Roberto Grasso",
"Giuseppe Curigliano",
"Gabriella Pravettoni"
],
"year": 2025,
"pdf_url": "https://www.frontiersin.org/journals/public-health/articles/10.3389/fpubh.2025.1520772/pdf",
"source": "openalex",
"doi": "https://doi.org/10.3389/fpubh.2025.1520772",
"abstract": "Equity in healthcare remains a pressing issue in cancer care across the European Union. Although numerous European initiatives address prevention, early diagnosis, and treatment, significant disparities in access to innovative cancer therapies persist. Time-to-reimbursement for new anticancer drugs varies widely between member states, depending on national health policies, economic capacity, and healthcare infrastructure. These differences particularly affect countries in Central and Eastern Europe, where delays in reimbursement, limited access to clinical trials, and restricted availability of specialized care contribute to worse outcomes. This narrative review examines how disparities in reimbursement timelines and access to new cancer therapies may affect factors such as early detection, specialized treatment availability, clinical trial participation, and socioeconomic status. The discussion is framed within the BEACON project, a European Union-funded initiative under the EU4Health programme. BEACON brings together patients, healthcare providers, researchers, and policymakers to create a cross-border network for quality-assured diagnosis and treatment. Through its multilingual digital platform, the project fosters collaboration, supports health literacy, and enhances access to innovative cancer therapies, aiming to reduce inequities regardless of geographic or socioeconomic background."
},
{
"venue": "ECIR",
"title": "Some Things Never Change: Overcoming Persistent Challenges in Children IR",
"authors": [
"Maria Soledad Pera",
"Theo Huibers",
"Emiliana Murgia",
"Monica Landoni"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730270",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730270",
"abstract": "There is a lack of a steady and solid influx of information retrieval (IR) research that has children (as the user group) as the protagonist. Existing work is scattered, conducted by only a few research groups, and often based on small-scale user studies or data that cannot be widely shared. Moreover, much of the current research focuses on specific age ranges and abilities, neglecting the broader spectrum of children's needs. Consequently, the paucity of IR research on how search and recommender systems serve and/or ultimately affect children translates into one of many 'Low-resource environments' in IR. Drawing from the literature and our experience in this area, we highlight key challenges and encourage greater attention from the IR community to address this critical gap."
},
{
"venue": "ECIR",
"title": "A Research of Challenges and Solutions in Retrieval Augmented Generation (RAG) Systems",
"authors": [
"Juan Gu"
],
"year": 2025,
"pdf_url": "https://drpress.org/ojs/index.php/HSET/article/download/28756/28231",
"source": "openalex",
"doi": "https://doi.org/10.54097/364hex16",
"abstract": "Retrieval-Augmented Generation (RAG) systems represent a significant innovation in the field of Natural Language Processing (NLP), ingeniously integrating Large Language Models (LLMs) with dynamic external knowledge retrieval. This amalgamation not only enhances the models' responsiveness to real-world knowledge but also addresses the limitations of conventional generative models in terms of knowledge update velocity and factual accuracy. This review examines the challenges faced by RAG systems and their solutions. It delves into the central architecture of RAG systems, encompassing retrieval components, generative components, and knowledge bases, with a particular focus on recent advancements that have expanded the boundaries of performance and functionality. The study critically analyzes major challenges such as retrieval efficiency and dynamic knowledge management. This paper evaluates various advanced solutions proposed in recent literature, comparing their efficacy and discussing the trade-offs involved. Ultimately, this paper aims to provide researchers, developers, and users of RAG systems with a comprehensive perspective, fostering ongoing innovation and the expansion of applications in this domain."
},
{
"venue": "ECIR",
"title": "Beyond Utility: Evaluating LLM as Recommender",
"authors": [
"Chumeng Jiang",
"Jiayin Wang",
"Weizhi Ma",
"Charles L. A. Clarke",
"Shuai Wang",
"Chuhan Wu",
"Min Zhang"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714759",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714759",
"abstract": "With the rapid development of Large Language Models (LLMs), recent studies employed LLMs as recommenders to provide personalized information services for distinct users. Despite efforts to improve the accuracy of LLM-based recommendation models, relatively little attention is paid to beyond-utility dimensions. Moreover, there are unique evaluation aspects of LLM-based recommendation models, which have been largely ignored. To bridge this gap, we explore four new evaluation dimensions and propose a multidimensional evaluation framework. The new evaluation dimensions include: 1) history length sensitivity, 2) candidate position bias, 3) generation-involved performance, and 4) hallucinations. All four dimensions have the potential to impact performance, but are largely unnecessary for consideration in traditional systems. Using this multidimensional evaluation framework, along with traditional aspects, we evaluate the performance of seven LLM-based recommenders, with three prompting strategies, comparing them with six traditional models on both ranking and re-ranking tasks on four datasets. We find that LLMs excel at handling tasks with prior knowledge and shorter input histories in the ranking setting, and perform better in the re-ranking setting, beating traditional models across multiple dimensions. However, LLMs exhibit substantial candidate position bias issues, and some models hallucinate nonexistent items much more often than others. We intend our evaluation framework and observations to benefit future research on the use of LLMs as recommenders. The code and data are available at https://github.com/JiangDeccc/EvaLLMasRecommender."
},
{
"venue": "ECIR",
"title": "Natural Language Processing for the Legal Domain: A Survey of Tasks, Datasets, Models, and Challenges",
"authors": [
"Farid Ariai",
"Joel Mackenzie",
"Gianluca Demartini"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3777009",
"source": "openalex",
"doi": "https://doi.org/10.1145/3777009",
"abstract": "Natural Language Processing (NLP) is revolutionising the way both professionals and laypersons operate in the legal field. The considerable potential for NLP in the legal sector, especially in developing computational assistance tools for various legal processes, has captured the interest of researchers for years. This survey follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses framework, reviewing 154 studies, with a final selection of 131 after manual filtering. It explores foundational concepts related to NLP in the legal domain, illustrating the unique aspects and challenges of processing legal texts, such as extensive document lengths, complex language, and limited open legal datasets. We provide an overview of NLP tasks specific to legal text, such as Document Summarisation, Named Entity Recognition, Question Answering, Argument Mining, Text Classification, and Judgment Prediction. Furthermore, we analyse both developed legal-oriented language models, and approaches for adapting general-purpose language models to the legal domain. Additionally, we identify sixteen open research challenges, including the detection and mitigation of bias in artificial intelligence applications, the need for more robust and interpretable models, and improving explainability to handle the complexities of legal language and reasoning."
},
{
"venue": "ECIR",
"title": "A Human-AI Comparative Analysis of Prompt Sensitivity in LLM-Based Relevance Judgment",
"authors": [
"Negar Arabzadeh",
"Charles L. A. Clarke"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730159",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730159",
"abstract": "Large Language Models (LLMs) are increasingly used to automate relevance judgments for information retrieval (IR) tasks, often demonstrating agreement with human labels that approaches inter-human agreement. To assess the robustness and reliability of LLM-based relevance judgments, we systematically investigate impact of prompt sensitivity on the task. We collected prompts for relevance assessment from 15 human experts and 15 LLMs across three tasks-binary, graded, and pairwise-yielding 90 prompts in total. We compare LLM-generated labels with TREC official human labels using Cohen's κ and pairwise agreement measures. In addition, we compare human- and LLM-generated prompts and analyze differences among different LLMs as judges. We release all data and prompts at https://github.com/Narabzad/prompt-sensitivity-relevance-judgements/."
},
{
"venue": "ECIR",
"title": "LLM-Assisted Relevance Assessments: When Should We Ask LLMs for Help?",
"authors": [
"Rikiya Takehi",
"Ellen M. Voorhees",
"Tetsuya Sakai",
"Ian Soboroff"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729916",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729916",
"abstract": "Test collections are information retrieval tools that allow researchers to quickly and easily evaluate ranking algorithms. While test collections have become an integral part of IR research, the process of data creation involves significant manual annotation effort, which often makes it very expensive and time-consuming. Consequently, test collections could become too small when the budget is limited, which may lead to unstable evaluations. As a cheaper alternative, recent studies have proposed the use of large language models (LLMs) to completely replace human assessors. However, while LLMs seem to somewhat correlate with human judgments, their predictions are not perfect and often show bias. Thus, a complete replacement with LLMs is argued to be too risky and not fully reliable."
},
{
"venue": "ECIR",
"title": "PEIR: Modeling Performance in Neural Information Retrieval",
"authors": [
"Pooya Khandel",
"Andrew Yates",
"Ana-Lucia Varbanescu",
"Maarten de Rijke",
"Andy D. Pimentel"
],
"year": 2025,
"pdf_url": "https://handle.uba.uva.nl/personal/pure/en/publications/peir-modeling-performance-in-neural-information-retrieval(c35327d5-00c3-4cc9-9d2b-dce62ee91804).html",
"source": "openalex",
"doi": "https://doi.org/10.1007/978-3-031-88711-6_18",
"abstract": ""
},
{
"venue": "ECIR",
"title": "Overview of QuantumCLEF 2025: The Second Quantum Computing Challenge for Information Retrieval and Recommender Systems at CLEF",
"authors": [
"Andrea Pasin",
"Maurizio Ferrari Dacrema",
"Washington Cunha",
"Marcos André Gonçalves",
"Paolo Cremonesi",
"Nicola Ferro"
],
"year": 2025,
"pdf_url": "https://re.public.polimi.it/bitstream/11311/1302008/1/overview-of-quantumclef-2025-the-second-quantum-computing-challenge-for-information-retrieval-and-recommender-systems-at-clef.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/978-3-032-04354-2_22",
"abstract": ""
},
{
"venue": "ECIR",
"title": "Neural Prioritisation for Web Crawling",
"authors": [
"Francesca Pezzuti",
"Sean MacAvaney",
"Nicola Tonellotto"
],
"year": 2025,
"pdf_url": "https://eprints.gla.ac.uk/view/author/60888.html>",
"source": "openalex",
"doi": "https://doi.org/10.1145/3731120.3744597",
"abstract": "Given the vast scale of the Web, crawling prioritisation techniques based on link graph traversal, popularity, link analysis, and textual content are frequently applied to surface documents that are most likely to be valuable. While existing techniques are effective for keyword-based search, both retrieval methods and user search behaviours are shifting from keyword-based matching to natural language semantic matching. The remarkable success of applying semantic matching and quality signals during ranking leads us to hypothesize that crawling could be improved by prioritizing Web pages with high semantic quality. To investigate this, we propose a semantic quality-driven prioritisation technique to enhance the effectiveness of crawling and align the crawler behaviour with recent shift towards natural language search. We embed semantic understanding directly into the crawling process -- leveraging recent neural semantic quality estimators to prioritise the crawling frontier -- with the goal of surfacing content that is semantically rich and valuable for modern search needs. Our experiments on the English subset of ClueWeb22-B and the Researchy Questions query set show that, compared to existing crawling techniques, neural crawling policies significantly improve harvest rate, maxNDCG, and search effectiveness during the early stages of crawling. Meanwhile, crawlers based on our proposed neural policies maintain comparable search performance on keyword queries from the MS MARCO Web Search query set. While this work does not propose a definitive and complete solution, it presents a forward-looking perspective on Web crawling and opens the door to a new line of research on leveraging semantic analysis to effectively align crawlers with the ongoing shift toward natural language search."
},
{
"venue": "ECIR",
"title": "A novel transformer-based dual attention architecture for the prediction of financial time series",
"authors": [
"Anita Hadizadeh",
"Mohammad Jafar Tarokh",
"Majid Mirzaee Ghazani"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s44443-025-00045-y.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s44443-025-00045-y",
"abstract": "Financial prediction has gained significant attention due to the complex and non-linear dynamics of the market. A promising approach for generating accurate predictions is Transformers. Encoder-decoder structures efficiently capture complex temporal dependencies and patterns within large-scale data. However, relying on a single attention mechanism may limit the model’s ability to capture more intricate relationships. This paper proposes a dual attention architecture to improve the encoder-decoder framework for financial forecasting. First, the Price Attention Network (PAN) extracts complex features from price data and forecasts future prices using historical price inputs. Two key improvements are introduced to enhance self-attention: a Masked Self-Attention module focusing on the most relevant information and Multi-head Attention facilitating more profound insights into the data. Second, the Nonprice Attention Network (NAN) is proposed as a parallel network that processes related financial features. This network utilizes ConvLSTM, BiGRU, and Self-Attention to dynamically weigh and extract meaningful information from nonprice data. Finally, the PAN and NAN networks are integrated, enhancing prediction accuracy. The proposed approach outperforms five state-of-the-art models. Moreover, qualitative assessments of over 26 financial datasets, spanning large and small datasets with short and long histories, further validate the proposed model's ability. Evaluations using seven metrics show the model’s superiority, achieving a Mean Absolute Error (MAE) of 0.01991, Mean Squared Error (MSE) of 0.00084, Mean Pinball Loss (MPL) of 0.00996, Symmetric Mean Absolute Percentage Error (SMAPE) of 3.03324, and Mean Absolute Scaled Error (MASE) of 1.85436. This framework represents a significant advancement in financial prediction, offering accurate and interpretable forecasts across various time series tasks."
},
{
"venue": "ECIR",
"title": "An Alternative to FLOPS Regularization to Effectively Productionize SPLADE-Doc",
"authors": [
"Aldo Porco",
"Dhruv Mehra",
"Igor Malioutov",
"Karthik Radhakrishnan",
"Moniba Keymanesh",
"Daniel Preoțiuc-Pietro",
"Sean MacAvaney",
"Pengxiang Cheng"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730163",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730163",
"abstract": "Learned Sparse Retrieval (LSR) models encode text as weighted term vectors, which need to be sparse to leverage inverted index structures during retrieval. SPLADE, the most popular LSR model, uses FLOPS regularization to encourage vector sparsity during training. However, FLOPS regularization does not ensure sparsity among terms-only within a given query or document. Terms with very high Document Frequencies (DFs) substantially increase latency in production retrieval engines, such as Apache Solr, due to their lengthy posting lists. To address the issue of high DFs, we present a new variant of FLOPS regularization: DF-FLOPS. This new regularization technique penalizes the usage of high-DF terms, thereby shortening posting lists and reducing retrieval latency. Unlike other inference-time sparsification methods, such as stopword removal, DF-FLOPS regularization allows for the selective inclusion of high-frequency terms in cases where the terms are truly salient. We find that DF-FLOPS successfully reduces the prevalence of high-DF terms and lowers retrieval latency (around 10x faster) in a production-grade engine while maintaining effectiveness both in-domain (only a 2.2-point drop in MRR@10) and cross-domain (improved performance in 12 out of 13 tasks on which we tested). With retrieval latencies on par with BM25, this work provides an important step towards making LSR practical for deployment in production-grade search engines."
},
{
"venue": "ECIR",
"title": "CG-RAG: Research Question Answering by Citation Graph Retrieval-Augmented LLMs",
"authors": [
"Yuntong Hu",
"Zhihan Lei",
"Z. G. Dai",
"Allen Zhang",
"Abhinav Angirekula",
"Zheng Zhang",
"Liang Zhao"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729920",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729920",
"abstract": "Research question answering requires accurate retrieval and contextual understanding of scientific literature. However, current Retrieval-Augmented Generation (RAG) methods often struggle to balance complex document relationships with precise information retrieval. In this paper, we introduce Contextualized Graph Retrieval-Augmented Generation (CG-RAG), a novel framework that integrates sparse and dense retrieval signals within graph structures to enhance retrieval efficiency and subsequently improve generation quality for research question answering. First, we propose a contextual graph representation for citation graphs, effectively capturing both explicit and implicit connections within and across documents. Next, we introduce Lexical-Semantic Graph Retrieval (LeSeGR), which seamlessly integrates sparse and dense retrieval signals with graph encoding. It bridges the gap between lexical precision and semantic understanding in citation graph retrieval, demonstrating generalizability to existing graph retrieval and hybrid retrieval methods. Finally, we present a context-aware generation strategy that utilizes the retrieved graph-structured information to generate precise and contextually enriched responses using large language models (LLMs). Extensive experiments on research question answering benchmarks across multiple domains demonstrate that our CG-RAG framework significantly outperforms RAG methods combined with various state-of-the-art retrieval approaches, delivering superior retrieval accuracy and generation quality."
},
{
"venue": "ECIR",
"title": "Classifying Citation Contexts using Reasoning Models",
"authors": [
"Larsen, Birger",
"Jurowetzki, Roman"
],
"year": 2025,
"pdf_url": "https://doi.org/10.6084/m9.figshare.30740519",
"source": "openalex",
"doi": "https://doi.org/10.6084/m9.figshare.30740519",
"abstract": "While some success has been achieved in the classification of citation contexts by using Large Language Models (LLMs), significant improvements are still needed to achieve sufficient accuracy for large-scale real-world applications. We propose to use recent LLM models with AI reasoning capabilities (like QwQ, DeepSeek R1 and Gemini Thinking) to study if reasoning processes have the potential to achieve such performance improvements. We outline how we plan to test this on two tasks: 1) classification of citation contexts, and 2) elimination of low-quality annotations from training sets. We present results of our experiments (extending the results presented at the SCOLIA workshop at ECIR’2025 - that significantly outperform published results on standard benchmarks), and we describe our experiences in creating a small annotated citation context dataset, and present some initial experiences of detecting low-quality annotations."
},
{
"venue": "ECIR",
"title": "Conversational Gold: Evaluating Personalized Conversational Search System Using Gold Nuggets",
"authors": [
"Zahra Abbasiantaeb",
"Simon Lupart",
"Leif Azzopardi",
"Jeff Dalton",
"Mohammad Aliannejadi"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730316",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730316",
"abstract": "The rise of personalized conversational search systems has been driven by advancements in Large Language Models (LLMs), enabling these systems to retrieve and generate answers for complex information needs. However, the automatic evaluation of responses generated by Retrieval Augmented Generation (RAG) systems remains an understudied challenge. In this paper, we introduce a new resource for assessing the retrieval effectiveness and relevance of responses generated by RAG systems, using a nugget-based evaluation framework. Built upon the foundation of TREC iKAT 2023, our dataset extends to the TREC iKAT 2024 collection, which includes 17 conversations and 20,575 relevance passage assessments, together with 2,279 extracted gold nuggets and 62 manually written gold answers from NIST assessors. While maintaining the core structure of its predecessor, this new collection enables a deeper exploration of generation tasks in conversational settings. Key improvements in iKAT 2024 include: (1) ''gold nuggets'' - concise, essential pieces of information extracted from relevant passages of the collection - which serve as a foundation for automatic response evaluation; (2) manually written answers to provide a gold standard for response evaluation; (3) expanded user personas, providing richer contextual grounding; and (4) a transition from Personal Text Knowledge Base (PTKB) ranking to PTKB classification and selection. Built on this resource, we provide a framework for long-form answer generation evaluation, involving nugget extraction and nugget matching, linked to retrieval. This establishes a solid resource for advancing research in personalized conversational search and long-form answer generation. Our resources are publicly available at https://github.com/irlabamsterdam/CONE-RAG."
},
{
"venue": "ECIR",
"title": "A Survey on LLM-powered Agents for Recommender Systems",
"authors": [
"Qiyao Peng",
"Hongtao Liu",
"Huang Hua",
"Jian Yang",
"Qing Yang",
"Minglai Shao"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.620.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.620",
"abstract": "Recently, Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language understanding, reasoning, and generation, prompting the recommendation community to leverage these powerful models to address fundamental challenges in traditional recommender systems, including limited comprehension of complex user intents, insufficient interaction capabilities, and inadequate recommendation interpretability.This survey presents a comprehensive synthesis of this rapidly evolving field.We consolidate existing studies into three paradigms: (i) recommenderoriented methods, which directly enhance core recommendation mechanisms; (ii) interactionoriented methods, which conduct multi-turn conversations to elicit preferences and deliver interpretable explanations; and (iii) simulationoriented methods, that model user-item interactions through multi-agent frameworks.Then, we dissect a four-module agent architecture: profile, memory, planning, and action.Then we review representative designs, public datasets, and evaluation protocols.Finally, we give the open challenges that impede real-world deployment, including cost-efficient inference, robust evaluation, and security."
},
{
"venue": "ECIR",
"title": "Surgical outcomes of robotic surgery for kidney stones: a systematic review and meta-analysis from section of YAU and EAU endourology",
"authors": [
"Rıfat Burak Ergül",
"Muhammet Firat Özervarlı",
"Frédéric Panthier",
"Carlotta Nedbal",
"Arman Tsaturyan",
"Amelia Pietropaolo",
"Giovanni Cacciamani",
"Peter Kronenberg",
"Pieter De Backer",
"Chady Ghnatios",
"Z. Khene",
"Sung Yong Cho",
"Ben Turney",
"Bhaskar Somani",
"Olivier Traxer",
"Tzevat Tefik"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s00345-025-05734-x.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s00345-025-05734-x",
"abstract": "BACKGROUND: Urinary tract stone disease is quite common in the population, and treatment modalities are constantly evolving. Technological advancements in endourology have led to a shift towards more minimally invasive treatments. Nowadays, robotic flexible ureteroscopy is becoming increasingly popular, showing promising results. This systematic review and meta-analysis aimed to evaluate the outcomes of robotic flexible ureteroscopy. MATERIALS AND METHODS: We conducted a systematic review in the PubMed, Scopus and Web of Science databases based on the 2020 Preferred Reporting Items for Systematic Review and Meta-Analyses guideline. Study protocol was registered at PROSPERO (CRD420251017383). All robotic flexible studies published until April 2025, which defined and provided the stone-free rate, were included. To assess surgical efficacy and reliability, stone size, operation time, and complications were also evaluated. Stone size was measured in a one-dimensional manner, based on the maximum length. RESULTS: A total of 320 studies were initially identified, with 11 full-text articles meeting the inclusion criteria, involving 656 patients and 660 renal units. The analysis included data from various robotic systems, including the Roboflex Avicenna, ILY, Senhance, and MONARCH platforms. The mean pooled stone-free rate was 86.0%, with a range from 57.7 to 96.5%, indicating variability across studies. The use of a random-effects model was justified by the presence of moderate-to-substantial heterogeneity across studies (I² = 63.5%, τ² = 0.627), and a statistically significant Q-test (p = 0.0022). The studies defined stone-free status as either complete stone clearance or residual fragments smaller than 2 mm. CONCLUSION: The analysis suggests that robotic URS is an effective and feasible treatment option for selected patients with urinary stones. Future research should focus on standardized reporting, comparative effectiveness studies, and cost-benefit analyses, while also addressing surgeon-centered outcomes such as ergonomic strain and musculoskeletal pain, to better define the role of robotic technology in endourological practice."
},
{
"venue": "ECIR",
"title": "Toward Holistic Evaluation of Recommender Systems Powered by Generative Models",
"authors": [
"Yashar Deldjoo",
"Nikhil Mehta",
"Maheswaran Sathiamoorthy",
"Shuai Zhang",
"Pablo Castells",
"Julian McAuley"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730354",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730354",
"abstract": "Recommender systems powered by generative models (Gen-RecSys) extend beyond classical item-ranking by producing open-ended content, which simultaneously unlocks richer user experiences and introduces new risks. On one hand, these systems can enhance personalization and appeal through dynamic explanations and multi-turn dialogues. On the other hand, they might venture into unknown territory-hallucinating nonexistent items, amplifying bias, or leaking private information. Traditional accuracy metrics cannot fully capture these challenges, as they fail to measure factual correctness, content safety, or alignment with user intent."
},
{
"venue": "ECIR",
"title": "The impact of monodisperse microbubble size on contrast-enhanced ultrasound super-localization imaging",
"authors": [
"Peiran Chen",
"Andreas Pollet",
"Simona Turco",
"Miguel de Vargas",
"Lisa Te Winkel",
"Wim van Hoeve",
"Jaap M. J. den Toonder",
"Hessel Wijkstra",
"Massimo Mischi"
],
"year": 2025,
"pdf_url": "https://pure.tue.nl/ws/files/356552395/2687_1_10.0036371.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1121/10.0036371",
"abstract": "Contrast-enhanced ultrasound (CEUS) super-localization imaging has shown promise for the assessment of microvascular networks by localizing and tracking microbubbles. The size of the available microbubbles for clinical use is polydisperse, but size-tailorable monodisperse microbubbles are now being developed that present a narrow size distribution. Therefore, proper frequency and pressure tuning have the potential to improve the signal-to-noise ratio and resolution of CEUS acquisitions, which can be expected to increase the performance of CEUS super-localization imaging. In this work, the impact of monodisperse microbubble size on CEUS imaging quality and the efficacy of super-localization imaging was investigated by jointly tuning different frequencies and pressures for different monodisperse microbubble size when performing in vitro CEUS imaging of microbubbles flowing through a dedicated sugar-printed dual-bifurcation microvasculature phantom. The obtained CEUS acquisitions were then post-processed to generate a super-localization output using the Gaussian-centroid localization approach. Four metrics, including generalized contrast-to-noise ratio, full-width half-maximum, number of localization events, and localization F1-score, were employed to quantify the CEUS imaging quality and super-localization performance. In general, jointly optimizing the transmit frequency and pressure for monodisperse microbubbles with smaller size leads to improved CEUS imaging and better super-localization performance. Yet, the weaker backscatter of smaller microbubbles must also be considered."
},
{
"venue": "ECIR",
"title": "Exploring the Effectiveness of Multi-stage Fine-Tuning for Cross-Encoder Re-rankers",
"authors": [
"Francesca Pezzuti",
"Sean MacAvaney",
"Nicola Tonellotto"
],
"year": 2025,
"pdf_url": "https://eprints.gla.ac.uk/view/author/60888.html>",
"source": "openalex",
"doi": "https://doi.org/10.1007/978-3-031-88714-7_45",
"abstract": ""
},
{
"venue": "ECIR",
"title": "Bridging the User-side Knowledge Gap in Knowledge-aware Recommendations with Large Language Models",
"authors": [
"Zheng Hu",
"Zhe Li",
"Ziyun Jiao",
"Satoshi Nakagawa",
"Jiawen Deng",
"Shi‐Min Cai",
"Tao Tang",
"Fuji Ren"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/33284/35439",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i11.33284",
"abstract": "In recent years, knowledge graphs have been integrated into recommender systems as item-side auxiliary information, enhancing recommendation accuracy. However, constructing and integrating structural user-side knowledge remains a significant challenge due to the improper granularity and inherent scarcity of user-side features. Recent advancements in Large Language Models (LLMs) offer the potential to bridge this gap by leveraging their human behavior understanding and extensive real-world knowledge. Nevertheless, integrating LLM-generated information into recommender systems presents challenges, including the risk of noisy information and the need for additional knowledge transfer. In this paper, we propose an LLM-based user-side knowledge inference method alongside a carefully designed recommendation framework to address these challenges. Our approach employs LLMs to infer user interests based on historical behaviors, integrating this user-side information with item-side and collaborative data to construct a hybrid structure: the Collaborative Interest Knowledge Graph (CIKG). Furthermore, we propose a CIKG-based recommendation framework that includes a user interest reconstruction module and a cross-domain contrastive learning module to mitigate potential noise and facilitate knowledge transfer. We conduct extensive experiments on three real-world datasets to validate the effectiveness of our method. Our approach achieves state-of-the-art performance compared to competitive baselines, particularly for users with sparse interactions."
},
{
"venue": "ECIR",
"title": "Large Language Model Relevance Assessors Agree With One Another More Than With Human Assessors",
"authors": [
"Maik Fröbe",
"Andrew Parry",
"Ferdinand Schlatt",
"Sean MacAvaney",
"Benno Stein",
"Martin Potthast",
"Matthias Hagen"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730218",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730218",
"abstract": "Relevance judgments can differ between assessors, but previous work has shown that such disagreements have little impact on the effectiveness rankings of retrieval systems. This applies to disagreements between humans as well as between human and large language model (LLM) assessors. However, the agreement between different LLM~assessors has not yet been systematically investigated. To close this gap, we compare eight LLM~assessors on the TREC DL tracks and the retrieval task of the RAG track with each other and with human assessors. We find that the agreement between LLM~assessors is higher than between LLMs and humans and, importantly, that LLM~assessors favor retrieval systems that use LLMs in their ranking decisions: our analyses with 30-50 retrieval systems show that the system rankings obtained by LLM~assessors overestimate LLM-based re-rankers by 9~to 17~positions on average."
},
{
"venue": "ECIR",
"title": "Overview of ROMCIR 2025: The 5thWorkshop on Reducing Online Misinformation through Credible Information Retrieval",
"authors": [
"Kruschwitz U.",
"Petrocchi M.",
"Viviani M."
],
"year": 2025,
"pdf_url": "https://ceur-ws.org/Vol-3986/",
"source": "openalex",
"doi": "",
"abstract": "ROMCIR 2025: The 5th Workshop on Reducing Online Misinformation through Credible Information Retrieval, is part of the Satellite Events of the 47th European Conference on Information Retrieval (ECIR 2025). TheWorkshop continues to serve as a key forum for advancing research and fostering dialogue on how to improve access to reliable information in an era marked by increasing information disorder. The challenge remains deeply complex, involving heterogeneous sources such as Websites, social media platforms, and multimedia content, across domains like misinformation detection, trustworthy Information Retrieval, propaganda mitigation, etc. In 2025, a growing focus is placed on understanding the dual role of generative technologies-particularly Large Language Models (LLMs)-in both unintentionally spreading misinformation and enhancing the capabilities of Information Retrieval Systems (IRSs). This year's program features keynote talks and peer-reviewed contributions that address critical topics including: the use of crowdsourcing to mitigate misinformation; the interplay between misinformation and LLMs, particularly in relation to fact-checking and rumor verification; and the societal implications of misinformation, with a special emphasis on its impact on children."
},
{
"venue": "ECIR",
"title": "Improving Low-Resource Retrieval Effectiveness Using Zero-Shot Linguistic Similarity Transfer",
"authors": [
"Andreas Chari",
"Sean MacAvaney",
"Iadh Ounis"
],
"year": 2025,
"pdf_url": "https://eprints.gla.ac.uk/343697/2/343697.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/978-3-031-88717-8_22",
"abstract": ""
},
{
"venue": "ECIR",
"title": "Unsupervised Corpus Poisoning Attacks in Continuous Space for Dense Retrieval",
"authors": [
"Yongkang Li",
"Panagiotis Eustratiadis",
"Simon Lupart",
"Evangelos Kanoulas"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730110",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730110",
"abstract": "This paper concerns corpus poisoning attacks in dense information retrieval, where an adversary attempts to compromise the ranking performance of a search algorithm by injecting a small number of maliciously generated documents into the corpus.Our work addresses two limitations in the current literature.First, attacks that perform adversarial gradient-based word substitution search do so in the discrete lexical space, while retrieval itself happens in the continuous embedding space.We thus propose an optimization method that operates in the embedding space directly.Specifically, we train a perturbation model with the objective of maintaining the geometric distance between the original and adversarial document embeddings, while also maximizing the token-level dissimilarity between the original and adversarial documents.Second, it is common for related work to have a strong assumption that the adversary has prior knowledge about the queries.In this paper, we focus on a more challenging variant of the problem where the adversary assumes no prior knowledge about the query distribution (hence, unsupervised).Our core contribution is an adversarial corpus attack that is fast and effective.We present comprehensive experimental results on both in-and out-of-domain datasets, focusing on two related tasks: a top-1 attack and a corpus poisoning attack.We consider attacks under both a white-box and a black-box setting.Notably, our method can generate successful adversarial examples in under two minutes per target document; four times faster compared to the fastest gradientbased word substitution methods in the literature with the same hardware.Furthermore, our adversarial generation method generates text that is more likely to occur under the distribution of natural text (low perplexity), and is therefore more difficult to detect."
},
{
"venue": "ECIR",
"title": "Investigations on segmentation-based fractal texture for texture classification in the presence of Gaussian noise",
"authors": [
"Shamik Tiwari",
"Akhilesh Sharma",
"Izzatdin Abdul Aziz",
"Deepak Gupta",
"Antima Jain",
"Hairulnizam Mahdin",
"Senthil Athithan",
"Rahmat Hidayat"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1371/journal.pone.0315135",
"source": "openalex",
"doi": "https://doi.org/10.1371/journal.pone.0315135",
"abstract": "Texture is a significant component used for several applications in content-based image retrieval. Any texture classification method aims to map an anonymously textured input image to one of the existing texture classes. Extensive ranges of methods for labeling image texture were proposed earlier. However, computing the performance of these methods in the presence of various degradations is always an open area of discussion. Image noise is always a dominant factor among various image degradation factors, affecting the performance of these methods and making texture classification challenging. Therefore, it is essential to investigate the interpretation of these methods in the presence of prominent degradation factors such as noise. Applications for Segmentation-Based Fractal Texture Features (SFTF) include image classification, texture generation, and medical image analysis. They are beneficial for examining textures with intricate, erratic patterns that are difficult to characterize using conventional statistical techniques accurately. This paper assesses two texture feature extraction methods based on SFTF and statistical moment-based texture features in the presence and absence of Gaussian noise. The SFTF and statistical moments-based handcrafted features are passed to a multilayer feed-forward neural network for classification. These models are evaluated on natural textures from Kylberg Texture Dataset 1.0. The results show the superiority of segmentation-based fractal analysis over other approaches. The average accuracy rates using the SFTF are 99% and 97% in the absence and presence of Gaussian noise, respectively."
},
{
"venue": "ECIR",
"title": "Axiomatic Re-Ranking for Argument Retrieval",
"authors": [
"Maximilian Heinrich",
"Marvin Vogel",
"Alexander Bondarenko",
"Matthias Hagen",
"Benno Stein"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730169",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730169",
"abstract": "Information retrieval axioms are formalized constraints that retrieval systems should ideally satisfy (e.g., to rank documents higher that contain the query terms more often). In this paper, we propose new axioms that focus on the scenario of argument retrieval: retrieval for queries that need arguments in the results. Our underlying axiomatic idea is that in such scenarios, documents should be prioritized with argumentative units that are similar to the query. We test our new axioms in re-ranking experiments on the data of the Touché ~2020 and~2021 shared task on argument retrieval for controversial questions, and show that the new axioms can improve the effectiveness of Touché's strong DirichletLM baseline model and even of the top-performing system from Touché ~2021, a system already specifically optimized for argument retrieval. Finally, we also propose a new method for visualizing the relationships between axioms based on their effects in re-ranking settings."
},
{
"venue": "ECIR",
"title": "TITE: Token-Independent Text Encoder for Information Retrieval",
"authors": [
"Ferdinand Schlatt",
"Thomas Hagen",
"Martin Potthast",
"Matthias Hagen"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730094",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730094",
"abstract": "Transformer-based retrieval approaches typically use the contextualized embedding of the first input token as a dense vector representation for queries and documents. The embeddings of all other tokens are also computed but then discarded, wasting resources. In this paper, we propose the Token-Independent Text Encoder (TITE) as a more efficient modification of the backbone encoder model. Using an attention-based pooling technique, TITE iteratively reduces the sequence length of hidden states layer by layer so that the final output is already a single sequence representation vector. Our empirical analyses on the TREC 2019 and 2020 Deep Learning tracks and the BEIR benchmark show that TITE is on par in terms of effectiveness compared to standard bi-encoder retrieval models while being up to 3.3 times faster at encoding queries and documents. Our code is available at: https://github.com/webis-de/SIGIR-25."
},
{
"venue": "ECIR",
"title": "Comprehensive Needs Assessment for Enhancing Self-Management in People with Lipoedema and the Support Provided by Their Healthcare Professionals",
"authors": [
"Lise Maren Kloosterman",
"Harriët Jager‐Wittenaar",
"Francine Schneider",
"Ad Hendrickx",
"Rienk Dekker",
"Aldo Scafoglieri"
],
"year": 2025,
"pdf_url": "https://www.dovepress.com/article/download/100616",
"source": "openalex",
"doi": "https://doi.org/10.2147/jmdh.s508816",
"abstract": "Background: The cause of lipoedema remains unclear, and the condition is currently incurable. Effective self-management is therefore essential for coping with its physical and psychological impacts and the necessary lifestyle adjustments. This study aimed to assess the needs, barriers, and facilitators for enhancing self-management and self-management support from the perspectives of people with lipoedema and the healthcare professionals (HCPs) involved in their care. Methods: The study used a mixed-methods approach, incorporating a narrative review focused on people with chronic conditions and their HCPs, along with focus groups involving people diagnosed with lipoedema and the HCPs involved in their care. The Core Processes of the Intervention Mapping method guided a systematic approach to address the study's objectives. Qualitative data were analyzed using a grounded theory approach. Results: Findings revealed unique self-management barriers for people with lipoedema, including limited awareness and expertise among HCPs, as well as stigmatization from both HCPs and society. Participants identified a need for tailored lifestyle plans, guidance, and support for monitoring progress. Key facilitators included self-management skills, supportive networks, and role models. HCPs noted barriers in communication and collaboration due to a lack of specialized professionals and negative attitudes toward lipoedema. They expressed a need for multidisciplinary/interprofessional teams, accurate diagnosis, patient openness, and reliable information resources. Facilitators included fostering trust, encouraging patient participation, and setting achievable goals. Conclusion: This study underscores the need for tailored self-management interventions for people with lipoedema. The adaptation of existing self-management strategies from other chronic conditions should take into account the specific needs, barriers, and facilitators of people with lipoedema and their HCPs."
},
{
"venue": "ECIR",
"title": "Multistakeholder fairness in tourism: what can algorithms learn from tourism management?",
"authors": [
"Peter Müllner",
"Anna Schreuer",
"Simone Kopeinik",
"Bernhard Wieser",
"Dominik Kowald"
],
"year": 2025,
"pdf_url": "https://www.frontiersin.org/journals/big-data/articles/10.3389/fdata.2025.1632766/pdf",
"source": "openalex",
"doi": "https://doi.org/10.3389/fdata.2025.1632766",
"abstract": "Algorithmic decision-support systems, i.e., recommender systems, are popular digital tools that help tourists decide which places and attractions to explore. However, algorithms often unintentionally direct tourist streams in a way that negatively affects the environment, local communities, or other stakeholders. This issue can be partly attributed to the computer science community's limited understanding of the complex relationships and trade-offs among stakeholders in the real world. In this work, we draw on the practical findings and methods from tourism management to inform research on multistakeholder fairness in algorithmic decision-support. Leveraging a semi-systematic literature review, we synthesize literature from tourism management as well as literature from computer science. Our findings suggest that tourism management actively tries to identify the specific needs of stakeholders and utilizes qualitative, inclusive and participatory methods to study fairness from a normative and holistic research perspective. In contrast, computer science lacks sufficient understanding of the stakeholder needs and primarily considers fairness through descriptive factors, such as measureable discrimination, while heavily relying on few mathematically formalized fairness criteria that fail to capture the multidimensional nature of fairness in tourism. With the results of this work, we aim to illustrate the shortcomings of purely algorithmic research and stress the potential and particular need for future interdisciplinary collaboration. We believe such a collaboration is a fundamental and necessary step to enhance algorithmic decision-support systems toward understanding and supporting true multistakeholder fairness in tourism."
},
{
"venue": "EMNLP",
"title": "A Survey of RAG-Reasoning Systems in Large Language Models",
"authors": [
"Yangning Li",
"Weizhi Zhang",
"Yuyao Yang",
"Wei‐Chieh Huang",
"Yu-Sian Wu",
"Junyu Luo",
"Yuanchen Bei",
"Henry Peng Zou",
"Xiao Ping Luo",
"Yusheng Zhao",
"Chunkit Chan",
"Yankai Chen",
"Zhongfen Deng",
"Yinghui Li",
"Hai-Tao Zheng",
"Dongyuan Li",
"Renhe Jiang",
"Ming Zhang",
"Yangqiu Song",
"Philip S. Yu"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.648.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.648",
"abstract": "Yangning Li, Weizhi Zhang, Yuyao Yang, Wei-Chieh Huang, Yaozu Wu, Junyu Luo, Yuanchen Bei, Henry Peng Zou, Xiao Luo, Yusheng Zhao, Chunkit Chan, Yankai Chen, Zhongfen Deng, Yinghui Li, Hai-Tao Zheng, Dongyuan Li, Renhe Jiang, Ming Zhang, Yangqiu Song, Philip S. Yu. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "Context Length Alone Hurts LLM Performance Despite Perfect Retrieval",
"authors": [
"Yufeng Du",
"Minyang Tian",
"Srikanth Ronanki",
"Subendhu Rongali",
"Sravan Bodapati",
"Aram Galstyan",
"Azton Wells",
"Roy Schwartz",
"E. A. Huerta",
"Hao Peng"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.1264.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.1264",
"abstract": "Yufeng Du, Minyang Tian, Srikanth Ronanki, Subendhu Rongali, Sravan Babu Bodapati, Aram Galstyan, Azton Wells, Roy Schwartz, Eliu A Huerta, Hao Peng. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "When Models Lie, We Learn: Multilingual Span-Level Hallucination Detection with PsiloQA",
"authors": [
"Elisei Rykov",
"Kseniia Petrushina",
"Maksim Savkin",
"Valerii Olisov",
"Artem Vazhentsev",
"Kseniia Titova",
"Alexander Panchenko",
"Vasily Konovalov",
"Julia Belikova"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.626.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.626",
"abstract": "Elisei Rykov, Kseniia Petrushina, Maksim Savkin, Valerii Olisov, Artem Vazhentsev, Kseniia Titova, Alexander Panchenko, Vasily Konovalov, Julia Belikova. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "Cache Saver: A Modular Framework for Efficient, Affordable, and Reproducible LLM Inference",
"authors": [
"Nearchos Potamitis",
"Lars Klein",
"Bardia Mohammadi",
"Chongyang Xu",
"Attreyee Mukherjee",
"Niket Tandon",
"Laurent Bindschaedler",
"Akhil Arora"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.1402.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.1402",
"abstract": "Nearchos Potamitis, Lars Henning Klein, Bardia Mohammadi, Chongyang Xu, Attreyee Mukherjee, Niket Tandon, Laurent Bindschaedler, Akhil Arora. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "DeCoRe: Decoding by Contrasting Retrieval Heads to Mitigate Hallucinations",
"authors": [
"Aryo Pradipta Gema",
"Jin Chen",
"Ahmed Abdulaal",
"Tom Diethe",
"Philip Teare",
"Beatrice Alex",
"Pasquale Minervini",
"Amrutha Saseendran"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.531.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.531",
"abstract": "Aryo Pradipta Gema, Chen Jin, Ahmed Abdulaal, Tom Diethe, Philip Alexander Teare, Beatrice Alex, Pasquale Minervini, Amrutha Saseendran. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "ClueAnchor: Clue-Anchored Knowledge Reasoning Exploration and Optimization for Retrieval-Augmented Generation",
"authors": [
"Hao Chen",
"Yukun Yan",
"Sen Mei",
"Wanxiang Che",
"Zhenghao Liu",
"Qi Shi",
"Xinze Li",
"Yuchun Fan",
"Pengcheng Huang",
"Qiushi Xiong",
"Zhiyuan Liu",
"Maosong Sun"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.1049.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.1049",
"abstract": "Hao Chen, Yukun Yan, Sen Mei, Wanxiang Che, Zhenghao Liu, Qi Shi, Xinze Li, Yuchun Fan, Pengcheng Huang, Qiushi Xiong, Zhiyuan Liu, Maosong Sun. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "A survey of multilingual large language models",
"authors": [
"Libo Qin",
"Qiguang Chen",
"Yuhang Zhou",
"Cheng Zhi",
"Yinghui Li",
"Lizi Liao",
"Min Li",
"Wanxiang Che",
"Philip S. Yu"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.patter.2024.101118",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.patter.2024.101118",
"abstract": "Multilingual large language models (MLLMs) leverage advanced large language models to process and respond to queries across multiple languages, achieving significant success in polyglot tasks. Despite these breakthroughs, a comprehensive survey summarizing existing approaches and recent developments remains absent. To this end, this paper presents a unified and thorough review of the field, highlighting recent progress and emerging trends in MLLM research. The contributions of this paper are as follows. (1) Extensive survey: to our knowledge, this is the pioneering thorough review of multilingual alignment in MLLMs. (2) Unified taxonomy: we provide a unified framework to summarize the current progress in MLLMs. (3) Emerging frontiers: key emerging frontiers are identified, alongside a discussion of associated challenges. (4) Abundant resources: we collect abundant open-source resources, including relevant papers, data corpora, and leaderboards. We hope our work can provide the community quick access and spur breakthrough research in MLLMs."
},
{
"venue": "EMNLP",
"title": "Benchmarking large language models for biomedical natural language processing applications and recommendations",
"authors": [
"Qingyu Chen",
"Yan Hu",
"Xueqing Peng",
"Qianqian Xie",
"Qiao Jin",
"Aidan Gilson",
"Maxwell Singer",
"X. C. Ai",
"Po-Ting Lai",
"Zhizheng Wang",
"Vipina K. Keloth",
"Kalpana Raja",
"Jimin Huang",
"Huan He",
"Fongci Lin",
"Jingcheng Du",
"Rui Zhang",
"W. Jim Zheng",
"Ron A. Adelman",
"Zhiyong Lu",
"Hua Xu"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41467-025-56989-2.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41467-025-56989-2",
"abstract": "The rapid growth of biomedical literature poses challenges for manual knowledge curation and synthesis. Biomedical Natural Language Processing (BioNLP) automates the process. While Large Language Models (LLMs) have shown promise in general domains, their effectiveness in BioNLP tasks remains unclear due to limited benchmarks and practical guidelines. We perform a systematic evaluation of four LLMs-GPT and LLaMA representatives-on 12 BioNLP benchmarks across six applications. We compare their zero-shot, few-shot, and fine-tuning performance with the traditional fine-tuning of BERT or BART models. We examine inconsistencies, missing information, hallucinations, and perform cost analysis. Here, we show that traditional fine-tuning outperforms zero- or few-shot LLMs in most tasks. However, closed-source LLMs like GPT-4 excel in reasoning-related tasks such as medical question answering. Open-source LLMs still require fine-tuning to close performance gaps. We find issues like missing information and hallucinations in LLM outputs. These results offer practical insights for applying LLMs in BioNLP."
},
{
"venue": "EMNLP",
"title": "Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inference",
"authors": [
"Benjamin C. Warner",
"Antoine Chaffin",
"Benjamin Clavié",
"Orion Weller",
"Oskar Hallström",
"Said Taghadouini",
"Alexis Gallagher",
"Raja Biswas",
"Faisal Ladhak",
"Tom Aarsen",
"Griffin Adams",
"Jeremy Howard",
"Iacopo Poli"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.acl-long.127.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.acl-long.127",
"abstract": "Benjamin Warner, Antoine Chaffin, Benjamin Clavié, Orion Weller, Oskar Hallström, Said Taghadouini, Alexis Gallagher, Raja Biswas, Faisal Ladhak, Tom Aarsen, Griffin Thomas Adams, Jeremy Howard, Iacopo Poli. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025."
},
{
"venue": "EMNLP",
"title": "Towards a holistic framework for multimodal LLM in 3D brain CT radiology report generation",
"authors": [
"Cheng-Yi Li",
"Kao-Jung Chang",
"Cheng-Fu Yang",
"Hsin‐Yu Wu",
"Wenting Chen",
"Hritik Bansal",
"Ling Chen",
"Yi‐Ping Yang",
"Yu‐Chun Chen",
"Shih‐Pin Chen",
"Shih‐Jen Chen",
"Jiing‐Feng Lirng",
"Kai-Wei Chang",
"Shih‐Hwa Chiou"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41467-025-57426-0.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41467-025-57426-0",
"abstract": "Multi-modal large language models (MLLMs) have transformed the landscape of modern healthcare, with automated radiology report generation (RRG) emerging as a cutting-edge application. While 2D MLLM-based RRG has been well established, its utility for 3D medical images remains largely unexplored. In this regard, we curate the 3D-BrainCT dataset (18,885 text-scan pairs) and develop BrainGPT, a clinically visual instruction-tuned (CVIT) model designed for 3D CT RRG. While we notice that the traditional LLM metrics failed to gauge the diagnostic quality of the RRG, we propose feature-oriented radiology task evaluation (FORTE), an evaluation scheme that captures the clinical essence of the generated reports. Here we show that BrainGPT achieves an average FORTE F1-score of 0.71 (degree = 0.661; landmark = 0.706; feature = 0.693, and impression = 0.779) and 74% of BrainGPT-generated reports were indistinguishable from human-written ground truth in a Turing-like test. Together, our work establishes a comprehensive framework encompassing dataset curation, anatomy-aware model fine-tuning, and the development of robust evaluation metrics for the RRG. By sharing our experience in 3D MLLM-based RRG, we aim to accelerate the expedition in human-machine collaboration for next-generation healthcare."
},
{
"venue": "EMNLP",
"title": "Evaluating large language model workflows in clinical decision support for triage and referral and diagnosis",
"authors": [
"Farieda Gaber",
"Maqsood Shaik",
"Fabio Allega",
"Agnes Julia Bilecz",
"Felix Busch",
"Kelsey Goon",
"Vedran Franke",
"Altuna Akalin"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41746-025-01684-1.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41746-025-01684-1",
"abstract": "Accurate medical decision-making is critical for both patients and clinicians. Patients often struggle to interpret their symptoms, determine their severity, and select the right specialist. Simultaneously, clinicians face challenges in integrating complex patient data to make timely, accurate diagnoses. Recent advances in large language models (LLMs) offer the potential to bridge this gap by supporting decision-making for both patients and healthcare providers. In this study, we benchmark multiple LLM versions and an LLM-based workflow incorporating retrieval-augmented generation (RAG) on a curated dataset of 2000 medical cases derived from the Medical Information Mart for Intensive Care database. Our findings show that these LLMs are capable of providing personalized insights into likely diagnoses, suggesting appropriate specialists, and assessing urgent care needs. These models may also support clinicians in refining diagnoses and decision-making, offering a promising approach to improving patient outcomes and streamlining healthcare delivery."
},
{
"venue": "EMNLP",
"title": "Leveraging long context in retrieval augmented language models for medical question answering",
"authors": [
"Gongbo Zhang",
"Zihan Xu",
"Qiao Jin",
"Fangyi Chen",
"Yilu Fang",
"Yi Liu",
"Justin F. Rousseau",
"Ziyang Xu",
"Zhiyong Lu",
"Chunhua Weng",
"Yifan Peng"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41746-025-01651-w.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41746-025-01651-w",
"abstract": "While holding great promise for improving and facilitating healthcare through applications of medical literature summarization, large language models (LLMs) struggle to produce up-to-date responses on evolving topics due to outdated knowledge or hallucination. Retrieval-augmented generation (RAG) is a pivotal innovation that improves the accuracy and relevance of LLM responses by integrating LLMs with a search engine and external sources of knowledge. However, the quality of RAG responses can be largely impacted by the rank and density of key information in the retrieval results, such as the \"lost-in-the-middle\" problem. In this work, we aim to improve the robustness and reliability of the RAG workflow in the medical domain. Specifically, we propose a map-reduce strategy, BriefContext, to combat the \"lost-in-the-middle\" issue without modifying the model weights. We demonstrated the advantage of the workflow with various LLM backbones and on multiple QA datasets. This method promises to improve the safety and reliability of LLMs deployed in healthcare domains by reducing the risk of misinformation, ensuring critical clinical content is retained in generated responses, and enabling more trustworthy use of LLMs in critical tasks such as medical question answering, clinical decision support, and patient-facing applications."
},
{
"venue": "EMNLP",
"title": "A Comprehensive Review of Multimodal Emotion Recognition: Techniques, Challenges, and Future Directions",
"authors": [
"You Wu",
"Qingwei Mi",
"Tianhan Gao"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2313-7673/10/7/418/pdf?version=1751272357",
"source": "openalex",
"doi": "https://doi.org/10.3390/biomimetics10070418",
"abstract": "This paper presents a comprehensive review of multimodal emotion recognition (MER), a process that integrates multiple data modalities such as speech, visual, and text to identify human emotions. Grounded in biomimetics, the survey frames MER as a bio-inspired sensing paradigm that emulates the way humans seamlessly fuse multisensory cues to communicate affect, thereby transferring principles from living systems to engineered solutions. By leveraging various modalities, MER systems offer a richer and more robust analysis of emotional states compared to unimodal approaches. The review covers the general structure of MER systems, feature extraction techniques, and multimodal information fusion strategies, highlighting key advancements and milestones. Additionally, it addresses the research challenges and open issues in MER, including lightweight models, cross-corpus generalizability, and the incorporation of additional modalities. The paper concludes by discussing future directions aimed at improving the accuracy, explainability, and practicality of MER systems for real-world applications."
},
{
"venue": "EMNLP",
"title": "Improving large language model applications in biomedicine with retrieval-augmented generation: a systematic review, meta-analysis, and clinical development guidelines",
"authors": [
"Siru Liu",
"Allison B. McCoy",
"Adam Wright"
],
"year": 2025,
"pdf_url": "https://academic.oup.com/jamia/advance-article-pdf/doi/10.1093/jamia/ocaf008/61442713/ocaf008.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1093/jamia/ocaf008",
"abstract": "OBJECTIVE: The objectives of this study are to synthesize findings from recent research of retrieval-augmented generation (RAG) and large language models (LLMs) in biomedicine and provide clinical development guidelines to improve effectiveness. MATERIALS AND METHODS: We conducted a systematic literature review and a meta-analysis. The report was created in adherence to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 analysis. Searches were performed in 3 databases (PubMed, Embase, PsycINFO) using terms related to \"retrieval augmented generation\" and \"large language model,\" for articles published in 2023 and 2024. We selected studies that compared baseline LLM performance with RAG performance. We developed a random-effect meta-analysis model, using odds ratio as the effect size. RESULTS: Among 335 studies, 20 were included in this literature review. The pooled effect size was 1.35, with a 95% confidence interval of 1.19-1.53, indicating a statistically significant effect (P = .001). We reported clinical tasks, baseline LLMs, retrieval sources and strategies, as well as evaluation methods. DISCUSSION: Building on our literature review, we developed Guidelines for Unified Implementation and Development of Enhanced LLM Applications with RAG in Clinical Settings to inform clinical applications using RAG. CONCLUSION: Overall, RAG implementation showed a 1.35 odds ratio increase in performance compared to baseline LLMs. Future research should focus on (1) system-level enhancement: the combination of RAG and agent, (2) knowledge-level enhancement: deep integration of knowledge into LLM, and (3) integration-level enhancement: integrating RAG systems within electronic health records."
},
{
"venue": "EMNLP",
"title": "Hallucination‐Free? Assessing the Reliability of Leading AI Legal Research Tools",
"authors": [
"Varun Magesh",
"Faiz Surani",
"Matthew Dahl",
"Mirac Suzgun",
"Christopher D. Manning",
"Daniel E. Ho"
],
"year": 2025,
"pdf_url": "https://onlinelibrary.wiley.com/doi/pdfdirect/10.1111/jels.12413",
"source": "openalex",
"doi": "https://doi.org/10.1111/jels.12413",
"abstract": "ABSTRACT Legal practice has witnessed a sharp rise in products incorporating artificial intelligence (AI). Such tools are designed to assist with a wide range of core legal tasks, from search and summarization of caselaw to document drafting. However, the large language models used in these tools are prone to “hallucinate,” or make up false information, making their use risky in high‐stakes domains. Recently, certain legal research providers have touted methods such as retrieval‐augmented generation (RAG) as “eliminating” or “avoid[ing]” hallucinations, or guaranteeing “hallucination‐free” legal citations. Because of the closed nature of these systems, systematically assessing these claims is challenging. In this article, we design and report on the first preregistered empirical evaluation of AI‐driven legal research tools. We demonstrate that the providers' claims are overstated. While hallucinations are reduced relative to general‐purpose chatbots (GPT‐4), we find that the AI research tools made by LexisNexis (Lexis+ AI) and Thomson Reuters (Westlaw AI‐Assisted Research and Ask Practical Law AI) each hallucinate between 17% and 33% of the time. We also document substantial differences between systems in responsiveness and accuracy. Our article makes four key contributions. It is the first to assess and report the performance of RAG‐based proprietary legal AI tools. Second, it introduces a comprehensive, preregistered dataset for identifying and understanding vulnerabilities in these systems. Third, it proposes a clear typology for differentiating between hallucinations and accurate legal responses. Last, it provides evidence to inform the responsibilities of legal professionals in supervising and verifying AI outputs, which remains a central open question for the responsible integration of AI into law."
},
{
"venue": "EMNLP",
"title": "Clinical entity augmented retrieval for clinical information extraction",
"authors": [
"Iván López",
"Akshay Swaminathan",
"Karthik S. Vedula",
"Sanjana Narayanan",
"Fateme Nateghi Haredasht",
"P. Stephen",
"April S. Liang",
"Steven Tate",
"Manoj Maddali",
"Robert J. Gallo",
"Nigam H. Shah",
"Jonathan H. Chen"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41746-024-01377-1.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41746-024-01377-1",
"abstract": "Large language models (LLMs) with retrieval-augmented generation (RAG) have improved information extraction over previous methods, yet their reliance on embeddings often leads to inefficient retrieval. We introduce CLinical Entity Augmented Retrieval (CLEAR), a RAG pipeline that retrieves information using entities. We compared CLEAR to embedding RAG and full-note approaches for extracting 18 variables using six LLMs across 20,000 clinical notes. Average F1 scores were 0.90, 0.86, and 0.79; inference times were 4.95, 17.41, and 20.08 s per note; average model queries were 1.68, 4.94, and 4.18 per note; and average input tokens were 1.1k, 3.8k, and 6.1k per note for CLEAR, embedding RAG, and full-note approaches, respectively. In conclusion, CLEAR utilizes clinical entities for information retrieval and achieves >70% reduction in token usage and inference time with improved performance compared to modern methods."
},
{
"venue": "EMNLP",
"title": "Scaling hermeneutics: a guide to qualitative coding with LLMs for reflexive content analysis",
"authors": [
"Zackary Okun Dunivin"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1140/epjds/s13688-025-00548-8.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1140/epjds/s13688-025-00548-8",
"abstract": "Abstract Qualitative coding, or content analysis, is more than just labeling text: it is a reflexive interpretive practice that shapes research questions, refines theoretical insights, and illuminates subtle social dynamics. As large language models (LLMs) become increasingly adept at nuanced language tasks, questions arise about whether—and how—they can assist in large-scale coding without eroding the interpretive depth that distinguishes qualitative analysis from traditional machine learning and other quantitative approaches to natural language processing. In this paper, we present a hybrid approach that preserves hermeneutic value while incorporating LLMs to scale the application of codes to large data sets that are impractical for manual coding. Our workflow retains the traditional cycle of codebook development and refinement, adding an iterative step to adapt definitions for machine comprehension, before ultimately replacing manual with automated text categorization. We demonstrate how to rewrite code descriptions for LLM-interpretation, as well as how structured prompts and prompting the model to explain its coding decisions (chain-of-thought) can substantially improve fidelity. Empirically, our case study of socio-historical codes highlights the promise of frontier AI language models to reliably interpret paragraph-long passages representative of a humanistic study. Throughout, we emphasize ethical and practical considerations, preserving space for critical reflection, and the ongoing need for human researchers’ interpretive leadership. These strategies can guide both traditional and computational scholars aiming to harness automation effectively and responsibly—maintaining the creative, reflexive rigor of qualitative coding while capitalizing on the efficiency afforded by LLMs."
},
{
"venue": "EMNLP",
"title": "Detecting AI-Generated Text: Factors Influencing Detectability with Current Methods",
"authors": [
"Kathleen Fraser",
"Hillary Dawkins",
"Svetlana Kiritchenko"
],
"year": 2025,
"pdf_url": "https://www.jair.org/index.php/jair/article/download/16665/27172",
"source": "openalex",
"doi": "https://doi.org/10.1613/jair.1.16665",
"abstract": "Large language models (LLMs) have advanced to a point that even humans have difficulty discerning whether a text was generated by another human, or by a computer. However, knowing whether a text was produced by human or artificial intelligence (AI) is important to determining its trustworthiness, and has applications in many domains including detecting fraud and academic dishonesty, as well as combating the spread of misinformation and political propaganda. The task of AI-generated text (AIGT) detection is therefore both very challenging, and highly critical. In this survey, we summarize stateof-the art approaches to AIGT detection, including watermarking, statistical and stylistic analysis, and machine learning classification. We also provide information about existing datasets for this task. Synthesizing the research findings, we aim to provide insight into the salient factors that combine to determine how “detectable” AIGT text is under different scenarios, and to make practical recommendations for future work towards this significant technical and societal challenge."
},
{
"venue": "EMNLP",
"title": "Understand What LLM Needs: Dual Preference Alignment for Retrieval-Augmented Generation",
"authors": [
"Guanting Dong",
"Yutao Zhu",
"Chenghao Zhang",
"Zechen Wang",
"Ji-Rong Wen",
"Zhicheng Dou"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714717",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714717",
"abstract": "Retrieval-augmented generation (RAG) has effectively mitigated the hallucination problem of large language models (LLMs). However, the difficulty of aligning the retriever with the LLMs' diverse knowledge preferences inevitably poses a challenge in developing a reliable RAG system. To address this issue, we propose DPA-RAG, a universal framework designed to align diverse knowledge preferences within RAG systems. Specifically, we initially introduce a preference knowledge construction pipeline and incorporate five novel query augmentation strategies to alleviate preference data scarcity. Based on preference data, DPA-RAG accomplishes both external and internal preference alignment: 1) It jointly integrates pairwise, pointwise, and contrastive preference alignment abilities into the reranker, achieving external preference alignment among RAG components. 2) It further introduces a pre-aligned stage before vanilla Supervised Fine-tuning (SFT), enabling LLMs to implicitly capture knowledge aligned with their reasoning preferences, achieving LLMs' internal alignment. Experimental results across four knowledge-intensive QA datasets demonstrate that DPA-RAG outperforms all baselines and seamlessly integrates both black-box and open-sourced LLM readers. Further qualitative analysis and discussions provide empirical guidance for achieving reliable RAG systems. Our code and example dataset are available at https://github.com/dongguanting/DPA-RAG."
},
{
"venue": "EMNLP",
"title": "Paths-over-Graph: Knowledge Graph Empowered Large Language Model Reasoning",
"authors": [
"Xingyu Tan",
"Xiaoyang Wang",
"Qing Liu",
"Xiwei Xu",
"Xin Yuan",
"Wenjie Zhang"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714892",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714892",
"abstract": "Large Language Models (LLMs) have achieved impressive results in various tasks but struggle with hallucination problems and lack of relevant knowledge, especially in deep complex reasoning and knowledge-intensive tasks.Knowledge Graphs (KGs), which capture vast amounts of facts in a structured format, offer a reliable source of knowledge for reasoning.However, existing KG-based LLM reasoning methods face challenges like handling multi-hop reasoning, multi-entity questions, and effectively utilizing graph structures.To address these issues, we propose Paths-over-Graph (PoG), a novel method that enhances LLM reasoning by integrating knowledge reasoning paths from KGs, improving the interpretability and faithfulness of LLM outputs.PoG tackles multi-hop and multi-entity questions through a three-phase dynamic multi-hop path exploration, which combines the inherent knowledge of LLMs with factual knowledge from KGs.In order to improve the efficiency, PoG prunes irrelevant information from the graph exploration first and introduces efficient three-step pruning techniques that incorporate graph structures, LLM prompting, and a pre-trained language model (e.g., SBERT) to effectively narrow down the explored candidate paths.This ensures all reasoning paths contain highly relevant information captured from KGs, making the reasoning faithful and interpretable in problem-solving.PoG innovatively utilizes graph structure to prune the irrelevant noise and represents the first method to implement multi-entity deep path detection on KGs for LLM reasoning tasks.Comprehensive experiments on five benchmark KGQA datasets demonstrate PoG outperforms the stateof-the-art method ToG across GPT-3.5-Turbo and GPT-4, achieving an average accuracy improvement of 18.9%.Notably, PoG with GPT-3.5-Turbosurpasses ToG with GPT-4 by up to 23.9%."
},
{
"venue": "EMNLP",
"title": "Large Language Models and Large Multimodal Models in Medical Imaging: A Primer for Physicians",
"authors": [
"Tyler Bradshaw",
"Xin Tie",
"Joshua Warner",
"Junjie Hu",
"Quanzheng Li",
"Xiang Li"
],
"year": 2025,
"pdf_url": "https://jnm.snmjournals.org/content/jnumed/early/2025/01/16/jnumed.124.268072.full.pdf",
"source": "openalex",
"doi": "https://doi.org/10.2967/jnumed.124.268072",
"abstract": "Large language models (LLMs) are poised to have a disruptive impact on health care. Numerous studies have demonstrated promising applications of LLMs in medical imaging, and this number will grow as LLMs further evolve into large multimodal models (LMMs) capable of processing both text and images. Given the substantial roles that LLMs and LMMs will have in health care, it is important for physicians to understand the underlying principles of these technologies so they can use them more effectively and responsibly and help guide their development. This article explains the key concepts behind the development and application of LLMs, including token embeddings, transformer networks, self-supervised pretraining, fine-tuning, and others. It also describes the technical process of creating LMMs and discusses use cases for both LLMs and LMMs in medical imaging."
},
{
"venue": "EMNLP",
"title": "LightRAG: Simple and Fast Retrieval-Augmented Generation",
"authors": [
"Zirui Guo",
"Lianghao Xia",
"Yanhua Yu",
"Tu Ao",
"Chao Huang"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.568.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.568",
"abstract": ""
},
{
"venue": "EMNLP",
"title": "A framework for evaluating the chemical knowledge and reasoning abilities of large language models against the expertise of chemists",
"authors": [
"A.H. Mirza",
"Nawaf Alampara",
"Sreekanth Kunchapu",
"Martiño Ríos-García",
"Benedict Emoekabu",
"Aswanth Krishnan",
"Tanya Gupta",
"Mara Schilling-Wilhelmi",
"Macjonathan Okereke",
"Anagha Aneesh",
"Mehrdad Asgari",
"J. Eberhardt",
"Amir Mohammad Elahi",
"Hani M. Elbeheiry",
"M.V. Gil",
"Christina Glaubitz",
"Maximilian Greiner",
"Caroline T. Holick",
"Tim Hoffmann",
"Abdelrahman Ibrahim",
"Lea C. Klepsch",
"Yannik Köster",
"Fabian Alexander Kreth",
"Jakob Meyer",
"Santiago Miret",
"Jan Matthias Peschel",
"Michael Ringleb",
"Nicole C. Roesner",
"J. Schreiber",
"Ulrich S. Schubert",
"Leanne M. Stafast",
"A. D. Dinga Wonanke",
"Michael Pieler",
"Philippe Schwaller",
"Kevin Maik Jablonka"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41557-025-01815-x.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41557-025-01815-x",
"abstract": "Large language models (LLMs) have gained widespread interest owing to their ability to process human language and perform tasks on which they have not been explicitly trained. However, we possess only a limited systematic understanding of the chemical capabilities of LLMs, which would be required to improve models and mitigate potential harm. Here we introduce ChemBench, an automated framework for evaluating the chemical knowledge and reasoning abilities of state-of-the-art LLMs against the expertise of chemists. We curated more than 2,700 question-answer pairs, evaluated leading open- and closed-source LLMs and found that the best models, on average, outperformed the best human chemists in our study. However, the models struggle with some basic tasks and provide overconfident predictions. These findings reveal LLMs' impressive chemical capabilities while emphasizing the need for further research to improve their safety and usefulness. They also suggest adapting chemistry education and show the value of benchmarking frameworks for evaluating LLMs in specific domains."
},
{
"venue": "EMNLP",
"title": "A clinically accessible small multimodal radiology model and evaluation metric for chest X-ray findings",
"authors": [
"Juan Manuel Zambrano Chaves",
"Shih-Cheng Huang",
"Yanbo Xu",
"Hanwen Xu",
"Naoto Usuyama",
"Sheng Zhang",
"Fei Wang",
"Yujia Xie",
"Mahmoud Khademi",
"Ziyi Yang",
"Hany Awadalla",
"Julia Gong",
"Houdong Hu",
"Jianwei Yang",
"Chunyuan Li",
"Jianfeng Gao",
"Yu Gu",
"Cliff Wong",
"Mu Wei",
"Tristan Naumann",
"Muhao Chen",
"Matthew P. Lungren",
"Akshay Chaudhari",
"Serena Yeung",
"Curtis P. Langlotz",
"Sheng Wang",
"Hoifung Poon"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41467-025-58344-x.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41467-025-58344-x",
"abstract": "Large foundation models show promise in biomedicine but face challenges in clinical use due to performance gaps, accessibility, cost, and lack of scalable evaluation. Here we show that open-source small multimodal models can bridge these gaps in radiology by generating free-text findings from chest X-ray images. Our data-centric approach leverages 697K curated radiology image-text pairs to train a specialized, domain-adapted chest X-ray encoder. We integrate this encoder with pre-trained language models via a lightweight adapter that aligns image and text modalities. To enable robust, clinically relevant evaluation, we develop and validate CheXprompt, a GPT-4-based metric for assessing factual accuracy aligned with radiologists' evaluations. Benchmarked with CheXprompt and other standard factuality metrics, LLaVA-Rad (7B) achieves state-of-the-art performance, outperforming much larger models like GPT-4V and Med-PaLM M (84B). While not immediately ready for real-time clinical deployment, LLaVA-Rad is a scalable, privacy-preserving and cost-effective step towards clinically adaptable multimodal AI for radiology."
},
{
"venue": "EMNLP",
"title": "TELEClass: Taxonomy Enrichment and LLM-Enhanced Hierarchical Text Classification with Minimal Supervision",
"authors": [
"Yunyi Zhang",
"Ruozhen Yang",
"Xueqiang Xu",
"Rui Li",
"Jinfeng Xiao",
"Jiaming Shen",
"Jiawei Han"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714940",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714940",
"abstract": "Hierarchical text classification aims to categorize each document into a set of classes in a label taxonomy, which is a fundamental web text mining task with broad applications such as web content analysis and semantic indexing. Most earlier works focus on fully or semi-supervised methods that require a large amount of human annotated data which is costly and time-consuming to acquire. To alleviate human efforts, in this paper, we work on hierarchical text classification with a minimal amount of supervision: using the sole class name of each node as the only supervision. Recently, large language models (LLM) have shown competitive performance on various tasks through zero-shot prompting, but this method performs poorly in the hierarchical setting because it is ineffective to include the large and structured label space in a prompt. On the other hand, previous weakly-supervised hierarchical text classification methods only utilize the raw taxonomy skeleton and ignore the rich information hidden in the text corpus that can serve as additional class-indicative features. To tackle the above challenges, we propose TELEClass, Taxonomy Enrichment and LLM-Enhanced weakly-supervised hierarchical text Classification, which combines the general knowledge of LLMs and task-specific features mined from an unlabeled corpus. TELEClass automatically enriches the raw taxonomy with class-indicative features for better label space understanding and utilizes novel LLM-based data annotation and generation methods specifically tailored for the hierarchical setting. Experiments show that TELEClass can significantly outperform previous baselines while achieving comparable performance to zero-shot prompting of LLMs with drastically less inference cost."
},
{
"venue": "EMNLP",
"title": "Retrieval-Augmented Generation (RAG)",
"authors": [
"Michael Klesel",
"H.F. Wittmann"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s12599-025-00945-3.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s12599-025-00945-3",
"abstract": ""
},
{
"venue": "EMNLP",
"title": "GRAG: Graph Retrieval-Augmented Generation",
"authors": [
"Yuntong Hu",
"Zhihan Lei",
"Zheng Zhang",
"Bo Pan",
"Chen Ling",
"Liang Zhao"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-naacl.232.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-naacl.232",
"abstract": "Naive Retrieval-Augmented Generation (RAG) focuses on individual documents during retrieval and, as a result, falls short in handling networked documents which are very popular in many applications such as citation graphs, social media, and knowledge graphs.To overcome this limitation, we introduce Graph Retrieval-Augmented Generation (GRAG), which tackles the fundamental challenges in retrieving textual subgraphs and integrating the joint textual and topological information into Large Language Models (LLMs) to enhance its generation.To enable efficient textual subgraph retrieval, we propose a novel divide-and-conquer strategy that retrieves the optimal subgraph structure in linear time.To achieve graph context-aware generation, incorporate textual graphs into LLMs through two complementary views-the text view and the graph view-enabling LLMs to more effectively comprehend and utilize the graph context.Extensive experiments on graph reasoning benchmarks demonstrate that in scenarios requiring multi-hop reasoning on textual graphs, our GRAG approach significantly outperforms current state-of-the-art RAG methods.Our datasets as well as codes of GRAG are available at https://github.com/HuieL/GRAG."
},
{
"venue": "EMNLP",
"title": "Retrieval-Augmented Generation (RAG) Chatbots for Education: A Survey of Applications",
"authors": [
"Jakub Swacha",
"Michał Gracel"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2076-3417/15/8/4234/pdf?version=1744376329",
"source": "openalex",
"doi": "https://doi.org/10.3390/app15084234",
"abstract": "Retrieval-Augmented Generation (RAG) overcomes the main barrier for the adoption of LLM-based chatbots in education: hallucinations. The uncomplicated architecture of RAG chatbots makes it relatively easy to implement chatbots that serve specific purposes and thus are capable of addressing various needs in the educational domain. With five years having passed since the introduction of RAG, the time has come to check the progress attained in its adoption in education. This paper identifies 47 papers dedicated to RAG chatbots’ uses for various kinds of educational purposes, which are analyzed in terms of their character, the target of the support provided by the chatbots, the thematic scope of the knowledge accessible via the chatbots, the underlying large language model, and the character of their evaluation."
},
{
"venue": "EMNLP",
"title": "GNN-RAG: Graph Neural Retrieval for Efficient Large Language Model Reasoning on Knowledge Graphs",
"authors": [
"Costas Mavromatis",
"George Karypis"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-acl.856.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-acl.856",
"abstract": "Retrieval-augmented generation (RAG) in Knowledge Graph Question Answering (KGQA) enhances the context of Large Language Models (LLMs) by incorporating information retrieved from the Knowledge Graph (KG).Most recent approaches rely on costly LLM calls to generate executable relation paths or traverse the KG, which is inefficient in complex KGQA tasks, such as those involving multi-hop or multi-entity questions.We introduce the GNN-RAG framework, which utilizes lightweight Graph Neural Networks (GNNs) for effective and efficient graph retrieval.The GNN learns to assign importance weights to nodes based on their relevance to the question, as well as the relevance of their neighboring nodes.This enables the framework to effectively handle context from distant nodes in the graph, improving retrieval performance.GNN-RAG retrieves the shortest paths connecting question entities to GNN answer candidates, providing this information as context for the LLM.Experimental results show that GNN-RAG achieves effective retrieval on two widely used KGQA benchmarks (WebQSP and CWQ), outperforming or matching GPT-4 performance with a 7B tuned LLM.Additionally, GNN-RAG excels on multi-hop and multi-entity questions outperforming LLM-based retrieval approaches by 8.9-15.5% points at answer F1.Furthermore, it surpasses long-context inference while using 9 fewer KG tokens."
},
{
"venue": "EMNLP",
"title": "Automatic Short Answer Grading in the LLM Era: Does GPT-4 with Prompt Engineering beat Traditional Models?",
"authors": [
"Rafael Ferreira Mello",
"Cleon Xavier",
"Luiz Rodrigues",
"Filipe Dwan Pereira",
"Luciano Cabral",
"Newarney Torrezão da Costa",
"Geber Ramalho",
"Dragan Gašević"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3706468.3706481",
"source": "openalex",
"doi": "https://doi.org/10.1145/3706468.3706481",
"abstract": "Assessing short answers in educational settings is challenging due to the need for scalability and accuracy, which led to the field of Automatic Short Answer Grading (ASAG). Traditional machine learning models, such as ensemble and embeddings, have been widely researched in ASAG, but they often suffer from generalizability issues. Recently, Large Language Models (LLMs) emerged as an alternative to optimize ASAG systems. However, previous research has failed to present a comprehensive analysis of LLMs' performance powered by prompt engineering strategies and compare its capabilities to traditional models. This study presents a comparative analysis between traditional machine learning models and GPT-4 in the context of ASAG. We investigated the effectiveness of different models and text representation techniques and explored prompt engineering strategies for LLMs. The results indicate that traditional machine learning models outperform LLMs. However, GPT-4 showed promising capabilities, especially when configured with optimized prompt components, such as few-shot examples and clear instructions. This study contributes to the literature by providing a detailed evaluation of LLM performance compared to traditional machine learning models in a multilingual ASAG context, offering insights for developing more efficient automatic grading systems."
},
{
"venue": "EMNLP",
"title": "FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts",
"authors": [
"Yichen Gong",
"Delong Ran",
"Jinyuan Liu",
"Conglei Wang",
"Tianshuo Cong",
"Anyu Wang",
"Sisi Duan",
"Xiaoyun Wang"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/34568/36723",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i22.34568",
"abstract": "Large Vision-Language Models (LVLMs) signify a groundbreaking paradigm shift within the Artificial Intelligence (AI) community, extending beyond the capabilities of Large Language Models (LLMs) by assimilating additional modalities (e.g., images). Despite this advancement, the safety of LVLMs remains adequately underexplored, with a potential overreliance on the safety assurances purported by their underlying LLMs. In this paper, we propose FigStep, a straightforward yet effective black-box jailbreak algorithm against LVLMs. Instead of feeding textual harmful instructions directly, FigStep converts the prohibited content into images through typography to bypass the safety alignment. The experimental results indicate that FigStep can achieve an average attack success rate of 82.50% on six promising open-source LVLMs. Not merely to demonstrate the efficacy of FigStep, we conduct comprehensive ablation studies and analyze the distribution of the semantic embeddings to uncover that the reason behind the success of FigStep is the deficiency of safety alignment for visual embeddings. Moreover, we compare FigStep with five text-only jailbreaks and four image-based jailbreaks to demonstrate the superiority of FigStep, i.e., negligible attack costs and better attack performance. Above all, our work reveals that current LVLMs are vulnerable to jailbreak attacks, which highlights the necessity of novel cross-modality safety alignment techniques."
},
{
"venue": "EMNLP",
"title": "Pushing The Limit of LLM Capacity for Text Classification",
"authors": [
"Yazhou Zhang",
"Mengyao Wang",
"Qiuchi Li",
"Prayag Tiwari",
"Jing Qin"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3701716.3715528",
"source": "openalex",
"doi": "https://doi.org/10.1145/3701716.3715528",
"abstract": "In this era of open-ended language modeling, where task boundaries are gradually fading, an urgent question emerges: have we made significant progress in text classification with the full benefit of LLMs? To answer this question, we propose RGPT, an adaptive boosting framework tailored to produce a specialized text classification LLM by recurrently ensembling a pool of base learners. The base learners are constructed by adaptively adjusting the distribution of training samples and iteratively fine-tuning LLMs with them. Such base learners are then ensembled to be a specialized text classification LLM, by recurrently incorporating the historical predictions from the previous learners. Through a comprehensive empirical comparison, we show that RGPT significantly outperforms 8 state-of-the-art (SoTA) PLMs and 7 SoTA LLMs on four benchmarks by 2.90% on average. Further evaluation experiments reveal a clear superiority of RGPT over average human classification performance."
},
{
"venue": "EMNLP",
"title": "Knowledge Graph-Guided Retrieval Augmented Generation",
"authors": [
"Xiangrong Zhu",
"Yuexiang Xie",
"Yi Liu",
"Yuanman Li",
"Wei Hu"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.naacl-long.449.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.naacl-long.449",
"abstract": "Xiangrong Zhu, Yuexiang Xie, Yi Liu, Yaliang Li, Wei Hu. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025."
},
{
"venue": "EMNLP",
"title": "TableBench: A Comprehensive and Complex Benchmark for Table Question Answering",
"authors": [
"Xianjie Wu",
"Jian Yang",
"Linzheng Chai",
"Ge Zhang",
"Jiaheng Liu",
"Xeron Du",
"Di Liang",
"D. Shu",
"Xianfu Cheng",
"T. Sun",
"Tongliang Li",
"Zhoujun Li",
"Guanglin Niu"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/34739/36894",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i24.34739",
"abstract": "Recent advancements in Large Language Models (LLMs) have markedly enhanced the interpretation and processing of tabular data, introducing previously unimaginable capabilities. Despite these achievements, LLMs still encounter significant challenges when applied in industrial scenarios, particularly due to the increased complexity of reasoning required with real-world tabular data, underscoring a notable disparity between academic benchmarks and practical applications. To address this discrepancy, we conduct a detailed investigation into the application of tabular data in industrial scenarios and propose a comprehensive and complex benchmark TableBench, including 18 fields within four major categories of table question answering (TableQA) capabilities. Furthermore, we introduce TableLLM, trained on our meticulously constructed training set TableInstruct, achieving comparable performance with GPT-3.5. Massive experiments conducted on TableBench indicate that both open-source and proprietary LLMs still have significant room for improvement to meet real-world demands, where the most advanced model, GPT-4, achieves only a modest score compared to humans."
},
{
"venue": "EMNLP",
"title": "Multiple large language models versus experienced physicians in diagnosing challenging cases with gastrointestinal symptoms",
"authors": [
"Xintian Yang",
"Tongxin Li",
"Han Wang",
"Rongchun Zhang",
"Zhi Ni",
"Na Liu",
"Huihong Zhai",
"Jianghai Zhao",
"Fandong Meng",
"Zhongyin Zhou",
"Shanhong Tang",
"Limei Wang",
"Xiangping Wang",
"Hui Luo",
"Gui Ren",
"Linhui Zhang",
"Xiaoyu Kang",
"Jun Wang",
"Bo Ning",
"Xiaoning Yang",
"Weijie Xue",
"Xiaoyin Zhang",
"Ning Chen",
"Rui Guo",
"Baiwen Li",
"Yajun Li",
"Yaling Liu",
"Tiantian Zhang",
"Shuhui Liang",
"Yong Lv",
"Yongzhan Nie",
"Daiming Fan",
"Lina Zhao",
"Yanglin Pan"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41746-025-01486-5.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41746-025-01486-5",
"abstract": "Faced with challenging cases, doctors are increasingly seeking diagnostic advice from large language models (LLMs). This study aims to compare the ability of LLMs and human physicians to diagnose challenging cases. An offline dataset of 67 challenging cases with primary gastrointestinal symptoms was used to solicit possible diagnoses from seven LLMs and 22 gastroenterologists. The diagnoses by Claude 3.5 Sonnet covered the highest proportion (95% confidence interval [CI]) of instructive diagnoses (76.1%, [70.6%-80.9%]), significantly surpassing all the gastroenterologists (p < 0.05 for all). Claude 3.5 Sonnet achieved a significantly higher coverage rate (95% CI) than that of the gastroenterologists using search engines or other traditional resource (76.1% [70.6%-80.9%] vs. 45.5% [40.7%-50.4%], p < 0.001). The study highlights that advanced LLMs may assist gastroenterologists with instructive, time-saving, and cost-effective diagnostic scopes in challenging cases."
},
{
"venue": "EMNLP",
"title": "A Survey of LLM-based Agents in Medicine: How far are we from Baymax?",
"authors": [
"Wenxuan Wang",
"Z. Ma",
"Zheng Wang",
"Wu Chenghan",
"Jiaming Ji",
"Wenting Chen",
"Xiang Li",
"Yixuan Yuan"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-acl.539.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-acl.539",
"abstract": "Large Language Models (LLMs) are transforming healthcare through the development of LLM-based agents that can understand, reason about, and assist with medical tasks.This survey provides a comprehensive review of LLM-based agents in medicine, examining their architectures, applications, and challenges.We analyze the key components of medical agent systems, including system profiles, clinical planning mechanisms, medical reasoning frameworks, and external capacity enhancement.The survey covers major application scenarios such as clinical decision support, medical documentation, training simulations, and healthcare service optimization.We discuss evaluation frameworks and metrics used to assess these agents' performance in healthcare settings.While LLM-based agents show promise in enhancing healthcare delivery, several challenges remain, including hallucination management, multimodal integration, implementation barriers, and ethical considerations.The survey concludes by highlighting future research directions, including advances in medical reasoning inspired by recent developments in LLM architectures, integration with physical systems, and improvements in training simulations.This work provides researchers and practitioners with a structured overview of the current state and future prospects of LLM-based agents in medicine."
},
{
"venue": "EMNLP",
"title": "Don't Do RAG: When Cache-Augmented Generation is All You Need for Knowledge Tasks",
"authors": [
"Brian J Chan",
"Chao-Ting Chen",
"Jui-Hung Cheng",
"Hen‐Hsen Huang"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3701716.3715490",
"source": "openalex",
"doi": "https://doi.org/10.1145/3701716.3715490",
"abstract": "Retrieval-augmented generation (RAG) has gained traction as a powerful approach for enhancing language models by integrating external knowledge sources. However, RAG introduces challenges such as retrieval latency, potential errors in document selection, and increased system complexity. With the advent of large language models (LLMs) featuring significantly extended context windows, this paper proposes an alternative paradigm, cache-augmented generation (CAG) that bypasses real-time retrieval. Our method involves preloading all relevant resources, especially when the documents or knowledge for retrieval are of a limited and manageable size, into the LLM's extended context and caching its runtime parameters. During inference, the model utilizes these preloaded parameters to answer queries without additional retrieval steps. Comparative analyses reveal that CAG eliminates retrieval latency and minimizes retrieval errors while maintaining context relevance. Performance evaluations across multiple benchmarks highlight scenarios where long-context LLMs either outperform or complement traditional RAG pipelines. These findings suggest that, for certain applications, particularly those with a constrained knowledge base, CAG provide a streamlined and efficient alternative to RAG, achieving comparable or superior results with reduced complexity."
},
{
"venue": "EMNLP",
"title": "A survey on multilingual large language models: corpora, alignment, and bias",
"authors": [
"Yuemei Xu",
"Ling Hu",
"Jiayi Zhao",
"Zihan Qiu",
"Kexin Xu",
"Yuqi Ye",
"Hanwen Gu"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s11704-024-40579-4.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s11704-024-40579-4",
"abstract": "Abstract Based on the foundation of Large Language Models (LLMs), Multilingual LLMs (MLLMs) have been developed to address the challenges faced in multilingual natural language processing, hoping to achieve knowledge transfer from high-resource languages to low-resource languages. However, significant limitations and challenges still exist, such as language imbalance, multilingual alignment, and inherent bias. In this paper, we aim to provide a comprehensive analysis of MLLMs, delving deeply into discussions surrounding these critical issues. First of all, we start by presenting an overview of MLLMs, covering their evolutions, key techniques, and multilingual capacities. Secondly, we explore the multilingual training corpora of MLLMs and the multilingual datasets oriented for downstream tasks that are crucial to enhance the cross-lingual capability of MLLMs. Thirdly, we survey the state-of-the-art studies of multilingual representations and investigate whether the current MLLMs can learn a universal language representation. Fourthly, we discuss bias on MLLMs, including its categories, evaluation metrics, and debiasing techniques. Finally, we discuss existing challenges and point out promising research directions of MLLMs."
},
{
"venue": "EMNLP",
"title": "Joint speech and text machine translation for up to 100 languages",
"authors": [
"SEAMLESS Communication Team",
"Loïc Barrault",
"Yu-An Chung",
"Mariano Coria Meglioli",
"David Dale",
"Ning Dong",
"Paul-Ambroise Duquenne",
"Hady Elsahar",
"Hongyu Gong",
"Kevin S. Heffernan",
"John P. Hoffman",
"Christopher Klaiber",
"Pengwei Li",
"Daniel J. Licht",
"Jean Maillard",
"Alice Rakotoarison",
"Kaushik Ram Sadagopan",
"Guillaume Wenzek",
"Ethan Ye",
"Bapi Akula",
"Peng‐Jen Chen",
"Naji El Hachem",
"Brian E. Ellis",
"Gabriel Mejia Gonzalez",
"Justin Haaheim",
"Prangthip Hansanti",
"Russ Howes",
"Bernie Huang",
"Min-Jae Hwang",
"Hirofumi Inaguma",
"Somya Jain",
"Elahe Kalbassi",
"Amanda Kallet",
"Ilia Kulikov",
"Janice Lam",
"Daniel Li",
"Xutai Ma",
"Ruslan Mavlyutov",
"Benjamin Peloquin",
"Mohamed M. Ramadan",
"Abinesh Ramakrishnan",
"Anna Sun",
"Kevin Tran",
"Tuan Tran",
"Igor Tufanov",
"Vish Vogeti",
"Carleigh Wood",
"Yilin Yang",
"Bokai Yu",
"Pierre Andrews",
"Can Balioglu",
"Marta R. Costa‐jussà",
"Onur Çelebi",
"Maha Elbayad",
"Cynthia Gao",
"Francisco Guzmán",
"Justine Kao",
"Ann Lee",
"Alexandre Mourachko",
"Juan Pino",
"Sravya Popuri",
"Christophe Ropers",
"Safiyyah Saleem",
"Holger Schwenk",
"Paden Tomasello",
"Changhan Wang",
"Jeff Wang",
"Skyler Wang"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41586-024-08359-z.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41586-024-08359-z",
"abstract": "remain underexplored. To address this gap, here we introduce SEAMLESSM4T-Massively Multilingual and Multimodal Machine Translation-a single model that supports speech-to-speech translation (101 to 36 languages), speech-to-text translation (from 101 to 96 languages), text-to-speech translation (from 96 to 36 languages), text-to-text translation (96 languages) and automatic speech recognition (96 languages). Built using a new multimodal corpus of automatically aligned speech translations and other publicly available data, SEAMLESSM4T is one of the first multilingual systems that can translate from and into English for both speech and text. Moreover, it outperforms the existing state-of-the-art cascaded systems, achieving up to 8% and 23% higher BLEU (Bilingual Evaluation Understudy) scores in speech-to-text and speech-to-speech tasks, respectively. Beyond quality, when tested for robustness, our system is, on average, approximately 50% more resilient against background noise and speaker variations in speech-to-text tasks than the previous state-of-the-art systems. We evaluated SEAMLESSM4T on added toxicity and gender bias to assess translation safety. For the former, we included two strategies for added toxicity mitigation working at either training or inference time. Finally, all contributions in this work are publicly available for non-commercial use to propel further research on inclusive speech translation technologies."
},
{
"venue": "EMNLP",
"title": "Retrieval-Augmented Generation (RAG) in Healthcare: A Comprehensive Review",
"authors": [
"Fnu Neha",
"Deepshikha Bhati",
"Deepak Kumar Shukla"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2673-2688/6/9/226/pdf?version=1757581602",
"source": "openalex",
"doi": "https://doi.org/10.3390/ai6090226",
"abstract": "Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) by integrating external knowledge retrieval to improve factual consistency and reduce hallucinations. Despite growing interest, its use in healthcare remains fragmented. This paper presents a Systematic Literature Review (SLR) following PRISMA guidelines, synthesizing 30 peer-reviewed studies on RAG in clinical domains, focusing on three of its most prevalent and promising applications in diagnostic support, electronic health record (EHR) summarization, and medical question answering. We synthesize the existing architectural variants (naïve, advanced, and modular) and examine their deployment across these applications. Persistent challenges are identified, including retrieval noise (irrelevant or low-quality retrieved information), domain shift (performance degradation when models are applied to data distributions different from their training set), generation latency, and limited explainability. Evaluation strategies are compared using both standard metrics and clinical-specific metrics, FactScore, RadGraph-F1, and MED-F1, which are particularly critical for ensuring factual accuracy, medical validity, and clinical relevance. This synthesis offers a domain-focused perspective to guide researchers, healthcare providers, and policymakers in developing reliable, interpretable, and clinically aligned AI systems, laying the groundwork for future innovation in RAG-based healthcare solutions."
},
{
"venue": "EMNLP",
"title": "Rankers, Judges, and Assistants: Towards Understanding the Interplay of LLMs in Information Retrieval Evaluation",
"authors": [
"Krisztian Balog",
"Donald Metzler",
"Zhen Qin"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730348",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730348",
"abstract": "Large language models (LLMs) are increasingly integral to information retrieval (IR), powering ranking, evaluation, and AI-assisted content creation. This widespread adoption necessitates a critical examination of potential biases arising from the interplay between these LLM-based components. This paper synthesizes existing research and presents novel experiment designs that explore how LLM-based rankers and assistants influence LLM-based judges. We provide the first empirical evidence of LLM judges exhibiting significant bias towards LLM-based rankers. Furthermore, we observe limitations in LLM judges' ability to discern subtle system performance differences. Contrary to some previous findings, our preliminary study does not find evidence of bias against AI-generated content. These results highlight the need for a more holistic view of the LLM-driven information ecosystem. To this end, we offer initial guidelines and a research agenda to ensure the reliable use of LLMs in IR evaluation."
},
{
"venue": "EMNLP",
"title": "Ask in Any Modality: A Comprehensive Survey on Multimodal Retrieval-Augmented Generation",
"authors": [
"Mohammad Mahdi Abootorabi",
"Amirhosein Zobeiri",
"Mahdi Dehghani",
"Mohammadali Mohammadkhani",
"Bardia Mohammadi",
"Omid Ghahroodi",
"Mahdieh Soleymani Baghshah",
"Ehsaneddin Asgari"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-acl.861.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-acl.861",
"abstract": "Mohammad Mahdi Abootorabi, Amirhosein Zobeiri, Mahdi Dehghani, Mohammadali Mohammadkhani, Bardia Mohammadi, Omid Ghahroodi, Mahdieh Soleymani Baghshah, Ehsaneddin Asgari. Findings of the Association for Computational Linguistics: ACL 2025. 2025."
},
{
"venue": "EMNLP",
"title": "DNABERT-S: pioneering species differentiation with species-aware DNA embeddings",
"authors": [
"Zhihan Zhou",
"Weimin Wu",
"Harrison Ho",
"Jiayi Wang",
"Lizhen Shi",
"Ramana V. Davuluri",
"Zhong Wang",
"Han Liu"
],
"year": 2025,
"pdf_url": "https://academic.oup.com/bioinformatics/article-pdf/41/Supplement_1/i255/63745376/btaf188.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1093/bioinformatics/btaf188",
"abstract": "SUMMARY: We introduce DNABERT-S, a tailored genome model that develops species-aware embeddings to naturally cluster and segregate DNA sequences of different species in the embedding space. Differentiating species from genomic sequences (i.e. DNA and RNA) is vital yet challenging, since many real-world species remain uncharacterized, lacking known genomes for reference. Embedding-based methods are therefore used to differentiate species in an unsupervised manner. DNABERT-S builds upon a pre-trained genome foundation model named DNABERT-2. To encourage effective embeddings to error-prone long-read DNA sequences, we introduce Manifold Instance Mixup (MI-Mix), a contrastive objective that mixes the hidden representations of DNA sequences at randomly selected layers and trains the model to recognize and differentiate these mixed proportions at the output layer. We further enhance it with the proposed Curriculum Contrastive Learning (C2LR) strategy. Empirical results on 28 diverse datasets show DNABERT-S's effectiveness, especially in realistic label-scarce scenarios. For example, it identifies twice more species from a mixture of unlabeled genomic sequences, doubles the Adjusted Rand Index (ARI) in species clustering, and outperforms the top baseline's performance in 10-shot species classification with just a 2-shot training. AVAILABILITY AND IMPLEMENTATION: Model, codes, and data are publically available at https://github.com/MAGICS-LAB/DNABERT_S."
},
{
"venue": "EMNLP",
"title": "Scientific hypothesis generation by large language models: laboratory validation in breast cancer treatment",
"authors": [
"Mohamed Abdel‐Rehim",
"Héctor Zenil",
"Oghenejokpeme I. Orhobor",
"Marie Fisher",
"Ross J. Collins",
"Elizabeth Bourne",
"Gareth W. Fearnley",
"Emma Tate",
"Holly X. Smith",
"Larisa Soldatova",
"Ross D. King"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1098/rsif.2024.0674",
"source": "openalex",
"doi": "https://doi.org/10.1098/rsif.2024.0674",
"abstract": "Large language models (LLMs) have transformed artificial intelligence (AI) and achieved breakthrough performance on a wide range of tasks. In science, the most interesting application of LLMs is for hypothesis formation. A feature of LLMs, which results from their probabilistic structure, is that the output text is not necessarily a valid inference from the training text. These are termed 'hallucinations', and are harmful in many applications. In science, some hallucinations may be useful: novel hypotheses whose validity may be tested by laboratory experiments. Here, we experimentally test the application of LLMs as a source of scientific hypotheses using the domain of breast cancer treatment. We applied the LLM GPT4 to hypothesize novel synergistic pairs of US Food and Drug Administration (FDA)-approved non-cancer drugs that target the MCF7 breast cancer cell line relative to the non-tumorigenic breast cell line MCF10A. In the first round of laboratory experiments, GPT4 succeeded in discovering three drug combinations (out of 12 tested) with synergy scores above the positive controls. GPT4 then generated new combinations based on its initial results, this generated three more combinations with positive synergy scores (out of four tested). We conclude that LLMs are a valuable source of scientific hypotheses."
},
{
"venue": "EMNLP",
"title": "Hyperedge overlap drives explosive transitions in systems with higher-order interactions",
"authors": [
"Federico Malizia",
"Santiago Lamata-Otín",
"Mattia Frasca",
"Vito Latora",
"Jesús Gómez‐Gardeñes"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41467-024-55506-1.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41467-024-55506-1",
"abstract": "Recent studies have shown that novel collective behaviors emerge in complex systems due to the presence of higher-order interactions. However, how the collective behavior of a system is influenced by the microscopic organization of its higher-order interactions is not fully understood. In this work, we introduce a way to quantify the overlap among the hyperedges of a higher-order network, and we show that real-world systems exhibit different levels of intra-order hyperedge overlap. We then study two types of dynamical processes on higher-order networks, namely complex contagion and synchronization, finding that intra-order hyperedge overlap plays a universal role in determining the collective behavior in a variety of systems. Our results demonstrate that the presence of higher-order interactions alone does not guarantee abrupt transitions. Rather, explosivity and bistability require a microscopic organization of the structure with a low value of intra-order hyperedge overlap."
},
{
"venue": "EMNLP",
"title": "Explainable AI chatbots towards XAI ChatGPT: A review",
"authors": [
"Attila Kővári"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.heliyon.2025.e42077",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.heliyon.2025.e42077",
"abstract": "Advances in artificial intelligence (AI) have had a major impact on natural language processing (NLP), even more so with the emergence of large-scale language models like ChatGPT. This paper aims to provide a critical review of explainable AI (XAI) methodologies for AI chatbots, with a particular focus on ChatGPT. Its main objectives are to investigate the applied methods that improve the explainability of AI chatbots, identify the challenges and limitations within them, and explore future research directions. Such goals emphasize the need for transparency and interpretability of AI systems to build trust with users and allow for accountability. While integrating such interdisciplinary methods, such as hybrid methods combining knowledge graphs with ChatGPT, enhancing explainability, they also highlight industry needs for explainability and user-centred design. This will be followed by a discussion of the balance between explainability and performance, then the role of human judgement, and finally the future of verifiable AI. These are the avenues through which insights can be used to guide the development of transparent, reliable and efficient AI chatbots."
},
{
"venue": "EMNLP",
"title": "Hierarchical graph-based integration network for propaganda detection in textual news articles on social media",
"authors": [
"Pir Noman Ahmad",
"Jiequn Guo",
"Nagwa M. AboElenein",
"Qazi Mazhar ul Haq",
"Sadique Ahmad",
"Abeer D. Algarni",
"Abdelhamied A. Ateya"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-024-74126-9.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-024-74126-9",
"abstract": "During the Covid-19 pandemic, the widespread use of social media platforms has facilitated the dissemination of information, fake news, and propaganda, serving as a vital source of self-reported symptoms related to Covid-19. Existing graph-based models, such as Graph Neural Networks (GNNs), have achieved notable success in Natural Language Processing (NLP). However, utilizing GNN-based models for propaganda detection remains challenging because of the challenges related to mining distinct word interactions and storing nonconsecutive and broad contextual data. In this study, we propose a Hierarchical Graph-based Integration Network (H-GIN) designed for detecting propaganda in text within a defined domain using multilabel classification. H-GIN is extracted to build a bi-layer graph inter-intra-channel, such as Residual-driven Enhancement and Processing (RDEP) and Attention-driven Multichannel feature Fusing (ADMF) with suitable labels at two distinct classification levels. First, RDEP procedures facilitate information interactions between distant nodes. Second, by employing these guidelines, ADMF standardizes the Tri-Channels 3-S (sequence, semantic, and syntactic) layer, enabling effective propaganda detection through related and unrelated information propagation of news representations into a classifier from the existing ProText, Qprop, and PTC datasets, thereby ensuring its availability to the public. The H-GIN model demonstrated exceptional performance, achieving an impressive 82% accuracy and surpassing current leading models. Notably, the model's capacity to identify previously unseen examples across diverse openness scenarios at 82% accuracy using the ProText dataset was particularly significant."
},
{
"venue": "EMNLP",
"title": "Debate on Graph: A Flexible and Reliable Reasoning Framework for Large Language Models",
"authors": [
"Ma Jie",
"Zhitao Gao",
"Qi Chai",
"Wangchun Sun",
"Pinghui Wang",
"Hongbin Pei",
"Jing Tao",
"Lingyun Song",
"Jun Liu",
"Chen Zhang",
"Lizhen Cui"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/34658/36813",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i23.34658",
"abstract": "Large Language Models (LLMs) may suffer from hallucinations in real-world applications due to the lack of relevant knowledge. In contrast, knowledge graphs encompass extensive, multi-relational structures that store a vast array of symbolic facts. Consequently, integrating LLMs with knowledge graphs has been extensively explored, with Knowledge Graph Question Answering (KGQA) serving as a critical touchstone for the integration. This task requires LLMs to answer natural language questions by retrieving relevant triples from knowledge graphs. However, existing methods face two significant challenges: *excessively long reasoning paths distracting from the answer generation*, and *false-positive relations hindering the path refinement*. In this paper, we propose an iterative interactive KGQA framework that leverages the interactive learning capabilities of LLMs to perform reasoning and Debating over Graphs (DoG). Specifically, DoG employs a subgraph-focusing mechanism, allowing LLMs to perform answer trying after each reasoning step, thereby mitigating the impact of lengthy reasoning paths. On the other hand, DoG utilizes a multi-role debate team to gradually simplify complex questions, reducing the influence of false-positive relations. This debate mechanism ensures the reliability of the reasoning process. Experimental results on five public datasets demonstrate the effectiveness and superiority of our architecture. Notably, DoG outperforms the state-of-the-art method ToG by 23.7% and 9.1% in accuracy on WebQuestions and GrailQA, respectively. Furthermore, the integration experiments with various LLMs on the mentioned datasets highlight the flexibility of DoG."
},
{
"venue": "EMNLP",
"title": "Enhancing Health Information Retrieval with RAG by prioritizing topical relevance and factual accuracy",
"authors": [
"Rishabh Upadhyay",
"Marco Viviani"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s10791-025-09505-5.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s10791-025-09505-5",
"abstract": "Abstract The exponential surge in online health information, coupled with its increasing use by non-experts, highlights the pressing need for advanced Health Information Retrieval (HIR) models that consider not only topical relevance but also the factual accuracy of the retrieved information, given the potential risks associated with health misinformation. To this aim, this paper introduces a solution driven by Retrieval-Augmented Generation (RAG), which leverages the capabilities of generative Large Language Models (LLMs) to enhance the retrieval of health-related documents grounded in scientific evidence. In particular, we propose a three-stage model: in the first stage, the user’s query is employed to retrieve topically relevant passages with associated references from a knowledge base constituted by scientific literature. In the second stage, these passages, alongside the initial query, are processed by LLMs to generate a contextually relevant rich text (GenText). In the last stage, the documents to be retrieved are evaluated and ranked both from the point of view of topical relevance and factual accuracy by means of their comparison with GenText, either through stance detection or semantic similarity. In addition to calculating factual accuracy, GenText can offer a layer of explainability for it, aiding users in understanding the reasoning behind the retrieval. Experimental evaluation of our model on benchmark datasets and against baseline models demonstrates its effectiveness in enhancing the retrieval of both topically relevant and factually accurate health information, thus presenting a significant step forward in the health misinformation mitigation problem."
},
{
"venue": "EMNLP",
"title": "SENTIMENT ANALYSIS IN SOCIAL MEDIA: HOW DATA SCIENCE IMPACTS PUBLIC OPINION KNOWLEDGE INTEGRATES NATURAL LANGUAGE PROCESSING (NLP) WITH ARTIFICIAL INTELLIGENCE (AI)",
"authors": [
"M. Shah Alam",
"Md Sabbir Hossain Mrida",
"Md. Atikur Rahman"
],
"year": 2025,
"pdf_url": "https://researchinnovationjournal.com/index.php/AJSRI/article/download/29/18",
"source": "openalex",
"doi": "https://doi.org/10.63125/r3sq6p80",
"abstract": "This systematic literature review investigates the advancements, methodologies, challenges, and application domains of sentiment analysis with a particular focus on informal digital text such as social media content. A total of 91 peer-reviewed articles published between 2010 and 2024 were carefully selected and analyzed using the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) framework to ensure methodological rigor, transparency, and reproducibility. The review spans traditional machine learning algorithms, deep learning models, and transformer-based architectures, examining their role in enhancing sentiment classification accuracy across various textual and multimodal inputs. Key themes emerging from the analysis include the evolution of multimodal sentiment analysis incorporating emojis, images, and videos; the growing focus on emotion classification beyond polarity detection; and the development of multilingual and cross-lingual sentiment systems that aim to extend sentiment mining beyond English-dominated datasets. Furthermore, a notable subset of studies addressed the complexities of detecting sarcasm, irony, and linguistic ambiguity, highlighting significant limitations in even the most advanced models. The review also discusses the growing body of research in financial, political, and health-related sentiment analysis, where domain-specific customization has proven critical for reliable prediction. Despite technical progress, challenges remain in areas such as data imbalance, inconsistent evaluation metrics, lack of cross-domain generalizability, and insufficient attention to ethical concerns, including algorithmic bias and explainability. This review contributes a synthesized and critical understanding of the current state of sentiment analysis and identifies key research gaps, offering a valuable reference point for scholars, developers, and practitioners aiming to improve the robustness, inclusivity, and ethical grounding of sentiment analysis systems."
},
{
"venue": "EMNLP",
"title": "SafetyPrompts: A Systematic Review of Open Datasets for Evaluating and Improving Large Language Model Safety",
"authors": [
"Paul Röttger",
"Fabio Pernisi",
"Bertie Vidgen",
"Dirk Hovy"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/34975/37130",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i26.34975",
"abstract": "The last two years have seen a rapid growth in concerns around the safety of large language models (LLMs). Researchers and practitioners have met these concerns by creating an abundance of datasets for evaluating and improving LLM safety. However, much of this work has happened in parallel, and with very different goals in mind, ranging from the mitigation of near-term risks around bias and toxic content generation to the assessment of longer-term catastrophic risk potential. This makes it difficult for researchers and practitioners to find the most relevant datasets for their use case, and to identify gaps in dataset coverage that future work may fill. To remedy these issues, we conduct a first systematic review of open datasets for evaluating and improving LLM safety. We review 144 datasets, which we identified through an iterative and community-driven process over the course of several months. We highlight patterns and trends, such as a trend towards fully synthetic datasets, as well as gaps in dataset coverage, such as a clear lack of non-English and naturalistic datasets. We also examine how LLM safety datasets are used in practice -- in LLM release publications and popular LLM benchmarks -- finding that current evaluation practices are highly idiosyncratic and make use of only a small fraction of available datasets. Our contributions are based on SafetyPrompts.com, a living catalogue of open datasets for LLM safety, which we plan to update continuously as the field of LLM safety develops."
},
{
"venue": "EMNLP",
"title": "SeaKR: Self-aware Knowledge Retrieval for Adaptive Retrieval Augmented Generation",
"authors": [
"Zijun Yao",
"Weijian Qi",
"Liangming Pan",
"Shulin Cao",
"Linmei Hu",
"Weichuan Liu",
"Lei Hou",
"Juanzi Li"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.acl-long.1312.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.acl-long.1312",
"abstract": "Zijun Yao, Weijian Qi, Liangming Pan, Shulin Cao, Linmei Hu, Liu Weichuan, Lei Hou, Juanzi Li. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025."
},
{
"venue": "EMNLP",
"title": "RadioRAG: Online Retrieval–Augmented Generation for Radiology Question Answering",
"authors": [
"Soroosh Tayebi Arasteh",
"Mahshad Lotfinia",
"Keno K. Bressem",
"Robert Siepmann",
"Lisa Adams",
"Dyke Ferber",
"Christiane Kühl",
"Jakob Nikolas Kather",
"Sven Nebelung",
"Daniel Truhn"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2407.15621",
"source": "openalex",
"doi": "https://doi.org/10.1148/ryai.240476",
"abstract": "© RSNA, 2025."
},
{
"venue": "EMNLP",
"title": "Agentic Retrieval-Augmented Generation: Advancing AI-Driven Information Retrieval and Processing",
"authors": [
"Abhai Pratap Singh",
"Adit Jamdar",
"Prerna Kaul"
],
"year": 2025,
"pdf_url": "https://www.ijcttjournal.org/2025/Volume-73%20Issue-1/IJCTT-V73I1P111.pdf",
"source": "openalex",
"doi": "https://doi.org/10.14445/22312803/ijctt-v73i1p111",
"abstract": "This paper explores the emerging field of Agentic Retrieval-Augmented Generation (Agentic RAG), an advanced approach to AI-driven information retrieval and processing. Building upon traditional Retrieval-Augmented Generation, Agentic RAG incorporates goal reasoning and self-direction, enabling AI systems to make informed decisions based on user context and intent. The study examines the fundamental components of Agentic RAG, including its multi-agent hierarchical architecture, key features, and enhancements over conventional systems. Applications across various domains, such as healthcare, financial services, businesses, and education, are discussed. The paper also addresses challenges in implementation, including mitigating AI hallucinations, ethical considerations, and computational scalability. Performance evaluation methods and metrics for Agentic RAG systems are outlined, along with case studies demonstrating their effectiveness. Finally, the paper explores future directions for research and development in this rapidly evolving field, highlighting its potential to revolutionize AI-driven information retrieval and processing."
},
{
"venue": "EMNLP",
"title": "Financial Analysis: Intelligent Financial Data Analysis System Based on LLM-RAG",
"authors": [
"Jingru Wang",
"Ding Wen",
"Xiaotong Zhu"
],
"year": 2025,
"pdf_url": "https://www.ewadirect.com/proceedings/ace/article/view/22221/pdf",
"source": "openalex",
"doi": "https://doi.org/10.54254/2755-2721/2025.22221",
"abstract": "In the modern financial sector, the exponential growth of data has made efficient and accurate financial data analysis increasingly crucial. Traditional methods, such as statistical analysis and rule-based systems, often struggle to process and derive meaningful insights from complex financial information effectively. These conventional approaches face inherent limitations in handling unstructured data, capturing intricate market patterns, and adapting to rapidly evolving financial contexts, resulting in reduced accuracy and delayed decision-making processes. To address these challenges, this paper presents an intelligent financial data analysis system that integrates Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG) technology. Our system incorporates three key components: a specialized preprocessing module for financial data standardization, an efficient vector-based storage and retrieval system, and a RAG-enhanced query processing module. Using the NASDAQ financial fundamentals dataset from 2010 to 2023, we conducted comprehensive experiments to evaluate system performance. Results demonstrate significant improvements across multiple metrics: the fully optimized configuration (gpt-3.5-turbo-1106+RAG) achieved 78.6% accuracy and 89.2% recall, surpassing the baseline model by 23 percentage points in accuracy while reducing response time by 34.8%. The system also showed enhanced efficiency in handling complex financial queries, though with a moderate increase in memory utilization. Our findings validate the effectiveness of integrating RAG technology with LLMs for financial analysis tasks and provide valuable insights for future developments in intelligent financial data processing systems."
},
{
"venue": "EMNLP",
"title": "Discovering sentiment insights: streamlining tourism review analysis with Large Language Models",
"authors": [
"Dario Guidotti",
"Laura Pandolfo",
"Luca Pulina"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s40558-024-00309-9.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s40558-024-00309-9",
"abstract": "With digital technology increasingly shaping the tourism industry, understanding customer sentiment and identifying key themes in reviews is crucial for enhancing service quality. However, traditional sentiment analysis and keyword extraction models typically demand significant time, computational resources, and labelled data for training. In this paper, we explore how Large Language Models (LLMs) can be leveraged to automatically classify reviews as positive or negative and extract relevant keywords without the need for dedicated training. Additionally, we frame the keyword extraction task as a tool to assist human users in comprehending and interpreting review content, especially in scenarios where ground truth labels for keywords are unavailable. To evaluate our approach, we conduct an experimental analysis using several datasets of tourism reviews and various LLMs. Our results demonstrate the reliability of LLMs as zero-shot classifiers for sentiment analysis and showcase the efficacy of the approach in extracting meaningful keywords from reviews, providing valuable insights into customer sentiments and preferences. Overall, this research contributes to the intersection of information technology and tourism by presenting a practical solution for sentiment analysis and keyword extraction in tourism reviews, leveraging the capabilities of LLMs as versatile tools for enhancing decision-making processes in the tourism industry."
},
{
"venue": "EMNLP",
"title": "Natural Language Processing Methods for Symbolic Music Generation and Information Retrieval: A Survey",
"authors": [
"Dinh-Viet-Toan Le",
"Louis Bigo",
"Dorien Herremans",
"Mikaela Keller"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3714457",
"source": "openalex",
"doi": "https://doi.org/10.1145/3714457",
"abstract": "Music is frequently associated with the notion of language, as both domains share several similarities, including the ability for their content to be represented as sequences of symbols. In computer science, the fields of Natural Language Processing (NLP) and Music Information Retrieval (MIR) reflect this analogy through a variety of similar tasks, such as author detection or content generation. This similarity has long encouraged the adaptation of NLP methods to process musical data, particularly symbolic music data, and the rise of Transformer neural networks has considerably strengthened this practice. This survey reviews NLP methods applied to symbolic music generation and information retrieval following two axes. We first propose an overview of representations of symbolic music inspired by text sequential representations. We then review a large set of computational models, particularly deep learning models, which have been adapted from NLP to process these musical representations for various MIR tasks. These models are described and categorized through different prisms with a highlight on their music-specialized mechanisms. We finally present a discussion surrounding the adequate use of NLP tools to process symbolic music data. This includes technical issues regarding NLP methods which may open several doors for further research into more effectively adapting NLP tools to symbolic MIR."
},
{
"venue": "EMNLP",
"title": "Open challenges and opportunities in federated foundation models towards biomedical healthcare",
"authors": [
"Xingyu Li",
"Peng Lu",
"Yu‐Ping Wang",
"Weihua Zhang"
],
"year": 2025,
"pdf_url": "https://biodatamining.biomedcentral.com/counter/pdf/10.1186/s13040-024-00414-9",
"source": "openalex",
"doi": "https://doi.org/10.1186/s13040-024-00414-9",
"abstract": "This survey explores the transformative impact of foundation models (FMs) in artificial intelligence, focusing on their integration with federated learning (FL) in biomedical research. Foundation models such as ChatGPT, LLaMa, and CLIP, which are trained on vast datasets through methods including unsupervised pretraining, self-supervised learning, instructed fine-tuning, and reinforcement learning from human feedback, represent significant advancements in machine learning. These models, with their ability to generate coherent text and realistic images, are crucial for biomedical applications that require processing diverse data forms such as clinical reports, diagnostic images, and multimodal patient interactions. The incorporation of FL with these sophisticated models presents a promising strategy to harness their analytical power while safeguarding the privacy of sensitive medical data. This approach not only enhances the capabilities of FMs in medical diagnostics and personalized treatment but also addresses critical concerns about data privacy and security in healthcare. This survey reviews the current applications of FMs in federated settings, underscores the challenges, and identifies future research directions including scaling FMs, managing data diversity, and enhancing communication efficiency within FL frameworks. The objective is to encourage further research into the combined potential of FMs and FL, laying the groundwork for healthcare innovations."
},
{
"venue": "EMNLP",
"title": "ShortGPT: Layers in Large Language Models are More Redundant Than You Expect",
"authors": [
"Xin Men",
"Mingyu Xu",
"Qingyu Zhang",
"Qianhao Yuan",
"Bingning Wang",
"Hongyu Lin",
"Yaojie Lu",
"Xianpei Han",
"Weipeng Chen"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-acl.1035.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-acl.1035",
"abstract": "As Large Language Models (LLMs) continue to advance, their computational overhead has increased significantly.In this study, we identify notable redundancy across the layers of LLMs, where some layers contribute minimally to the overall network functionality.To quantify this, we introduce a metric called Block Influence (BI), which measures the importance of each layer based on the similarity between its input and output.Based on the observation of layer redundancy, we propose straightforward pruning methods for different tasks: ShortGPT for multiple-choice tasks and ShortGPT-gen for generative tasks.They prune redundant layers based on their BI scores.Our methods demonstrate superior performance over previous pruning methods.The ability to achieve better results through simple layer pruning, as opposed to more complex pruning techniques, suggests a high degree of redundancy across layers.We hope this work will contribute to future research for improving LLM efficiency."
},
{
"venue": "EMNLP",
"title": "How well can a large language model explain business processes as perceived by users?",
"authors": [
"Dirk Fahland",
"Fabiana Fournier",
"Lior Limonad",
"Inna Skarbovsky",
"Ava Swevels"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.datak.2025.102416",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.datak.2025.102416",
"abstract": "Large Language Models (LLMs) are trained on a vast amount of text to interpret and generate human-like textual content. They are becoming a vital vehicle in realizing the vision of the autonomous enterprise, with organizations today actively adopting LLMs to automate many aspects of their operations. LLMs are likely to play a prominent role in future AI-augmented business process management systems (ABPMSs) catering functionalities across all system lifecycle stages. One such system’s functionality is Situation-Aware eXplainability (SAX), which relates to generating causally sound and yet human-interpretable explanations that take into account the process context in which the explained condition occurred. In this paper, we present the SAX4BPM framework developed to generate SAX explanations. The SAX4BPM suite consists of a set of services and a central knowledge repository. The functionality of these services is to elicit the various knowledge ingredients that underlie SAX explanations. A key innovative component among these ingredients is the causal process execution view. In this work, we integrate the framework with an LLM to leverage its power to synthesize the various input ingredients for the sake of improved SAX explanations. Since the use of LLMs for SAX is also accompanied by a certain degree of doubt related to its capacity to adequately fulfill SAX along with its tendency for hallucination and lack of inherent capacity to reason, we pursued a methodological evaluation of the perceived quality of the generated explanations. To this aim, we developed a designated scale and conducted a rigorous user study. Our findings show that the input presented to the LLMs aided with the guard-railing of its performance, yielding SAX explanations having better-perceived fidelity. This improvement is moderated by the perception of trust and curiosity. More so, this improvement comes at the cost of the perceived interpretability of the explanation. • How can an LLM generate high-quality explanations for different process outcomes? • How can process knowledge be integrated into a user-interpretable explanation? • How can explanations reflect the process’s timely unfolding as perceived by users? • How can explanations accurately convey the process’s chain of dependencies?"
},
{
"venue": "EMNLP",
"title": "Harnessing Large Language Models for Knowledge Graph Question Answering via Adaptive Multi-Aspect Retrieval-Augmentation",
"authors": [
"Derong Xu",
"Xinhang Li",
"Ziheng Zhang",
"Zhenxi Lin",
"Zhihong Zhu",
"Zhi Zheng",
"Xian Wu",
"Xiangyu Zhao",
"Tong Xu",
"Enhong Chen"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/34747/36902",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i24.34747",
"abstract": "Large Language Models (LLMs) demonstrate remarkable capabilities, yet struggle with hallucination and outdated knowledge when tasked with complex knowledge reasoning, resulting in factually incorrect outputs. Previous studies have attempted to mitigate it by retrieving factual knowledge from large-scale knowledge graphs (KGs) to assist LLMs in logical reasoning and prediction of answers. However, this kind of approach often introduces noise and irrelevant data, especially in situations with extensive context from multiple knowledge aspects. In this way, LLM attention can be potentially mislead from question and relevant information. In our study, we introduce an Adaptive Multi-Aspect Retrieval-augmented over KGs (Amar) framework. This method retrieves knowledge including entities, relations, and subgraphs, and converts each piece of retrieved text into prompt embeddings. The Amar framework comprises two key sub-components: 1) a self-alignment module that aligns commonalities among entities, relations, and subgraphs to enhance retrieved text, thereby reducing noise interference; 2) a relevance gating module that employs a soft gate to learn the relevance score between question and multi-aspect retrieved data, to determine which information should be used to enhance LLMs' output, or even filtered altogether. Our method has achieved state-of-the-art performance on two common datasets, WebQSP and CWQ, showing a 1.9% improvement in accuracy over its best competitor and a 6.6% improvement in logical form generation over a method that directly uses retrieved text as context prompts. These results demonstrate the effectiveness of Amar in improving the reasoning of LLMs."
},
{
"venue": "EMNLP",
"title": "AI-assisted Living Evidence Databases for Conservation Science",
"authors": [
"Sadiq Jaffer",
"William H. Morgan",
"S.A. Reynolds",
"Alec Christie",
"Anil Madhavapeddy",
"William J. Sutherland"
],
"year": 2025,
"pdf_url": "https://www.cambridge.org/engage/api-gateway/coe/assets/orp/resource/item/68e52a47bc2ac3a0e0c0eda8/original/ai-assisted-living-evidence-databases-for-conservation-science.pdf",
"source": "openalex",
"doi": "https://doi.org/10.33774/coe-2025-rmsqf",
"abstract": "Living evidence databases offer a robust and dynamic alternative to static systematic reviews but require a resilient technical infrastructure for continuous evidence processing. This working paper describes the architecture and implementation of a complete, end-to-end pipeline for this purpose, developed initially for the conservation science domain. Designed to operate on local infrastructure using self-hosted models, the system ingests and normalizes documents from academic publishers, screens them for relevance using a multi-stage process, and extracts structured data according to a predefined schema. Key features include a hybrid retrieval model; a human-AI collaborative process for refining inclusion criteria from complex protocols, and the integration of an established, statistically-principled stopping rule to ensure efficiency. In a baseline evaluation against a prior large-scale manual review, the fully automated pipeline achieved 97% recall and identified a significant number of relevant studies not included in the original review, demonstrating its viability as a foundational tool for maintaining living evidence databases."
},
{
"venue": "EMNLP",
"title": "Dual retrieving and ranking medical large language model with retrieval augmented generation",
"authors": [
"Qimin Yang",
"Huan Zuo",
"R. Su",
"H. T. Su",
"Tangyi Zeng",
"Huimei Zhou",
"Rongsheng Wang",
"Jiexin Chen",
"Yijun Lin",
"Zhiyi Chen",
"Tao Tan"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-00724-w.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-00724-w",
"abstract": "Recent advancements in large language models (LLMs) have significantly enhanced text generation across various sectors; however, their medical application faces critical challenges regarding both accuracy and real-time responsiveness. To address these dual challenges, we propose a novel two-step retrieval and ranking retrieval-augmented generation (RAG) framework that synergistically combines embedding search with Elasticsearch technology. Built upon a dynamically updated medical knowledge base incorporating expert-reviewed documents from leading healthcare institutions, our hybrid architecture employs ColBERTv2 for context-aware result ranking while maintaining computational efficiency. Experimental results show a 10% improvement in accuracy for complex medical queries compared to standalone LLM and single-search RAG variants, while acknowledging that latency challenges remain in emergency situations requiring sub-second responses in an experimental setting, which can be achieved in real-time using more powerful hardware in real-world deployments. This work establishes a new paradigm for reliable medical AI assistants that successfully balances accuracy and practical deployment considerations."
},
{
"venue": "EMNLP",
"title": "LLM-Based Agents for Tool Learning: A Survey",
"authors": [
"Weikai Xu",
"Chengrui Huang",
"Shen Gao",
"Shuo Shang"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s41019-025-00296-9.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s41019-025-00296-9",
"abstract": "Abstract Human beings capable of making and using tools can accomplish tasks far beyond their innate abilities, and this paradigm of integration with tools may not be limited to humans themselves. Recently, the large language model (LLM) has demonstrated immense potential across various fields with its unique planning and reasoning abilities. However, there are still many challenges beyond its capabilities due to deficiencies in its training data and inherent illusions. Thus, integrating LLMs and tools into tool learning agents has become a new emerging research direction. To this end, we present a systematic investigation and comprehensive review of tool-learning agents in this paper. We start by introducing the definition of the tool learning task for Agents and then illustrating the typical architecture of the tool-learning models. Since these tools are all defined by users, LLM does not know what tools there are and what their functions are. Thus, LLMs should first find appropriate tools and split the tool retrieval methods into two categories: training-based and non-training-based. To accurately complete the user task, it is important to decompose the task into several sub-tasks and execute them in the correct order. Following that, we introduce the tool planning methods and organize these works by whether they rely on the model’s inherent reasoning capabilities for planning or utilize external reasoning tools. Due to the rapid development of this field, we also introduce an emerging frontier direction: using multimodal tools for LLM. In addition, we compile current open-source benchmarks and evaluation metrics, focusing on their scale, composition, calculation methods, and assessment dimensions. Next, we introduce several application scenarios for the LLM-based tool learning methods. Finally, we discuss the safety and ethical issues involved in tool learning."
},
{
"venue": "EMNLP",
"title": "Retrieval-augmented generation elevates local LLM quality in radiology contrast media consultation",
"authors": [
"Akihiko Wada",
"Yuya Tanaka",
"Mitsuo Nishizawa",
"Akira Yamamoto",
"Toshiaki Akashi",
"Akifumi Hagiwara",
"Yayoi Hayakawa",
"Junko Kikuta",
"Keigo Shimoji",
"Katsuhiro Sano",
"Koji Kamagata",
"Atsushi Nakanishi",
"Shigeki Aoki"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41746-025-01802-z.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41746-025-01802-z",
"abstract": "Large language models (LLMs) demonstrate significant potential in healthcare applications, but clinical deployment is limited by privacy concerns and insufficient medical domain training. This study investigated whether retrieval-augmented generation (RAG) can improve locally deployable LLM for radiology contrast media consultation. In 100 synthetic iodinated contrast media consultations we compared Llama 3.2-11B (baseline and RAG) with three cloud-based models-GPT-4o mini, Gemini 2.0 Flash and Claude 3.5 Haiku. A blinded radiologist ranked the five replies per case, and three LLM-based judges scored accuracy, safety, structure, tone, applicability and latency. Under controlled conditions, RAG eliminated hallucinations (0% vs 8%; χ²₍Yates₎ = 6.38, p = 0.012) and improved mean rank by 1.3 (Z = -4.82, p < 0.001), though performance gaps with cloud models persist. The RAG-enhanced model remained faster (2.6 s vs 4.9-7.3 s) while the LLM-based judges preferred it over GPT-4o mini, though the radiologist ranked GPT-4o mini higher. RAG thus provides meaningful improvements for local clinical LLMs while maintaining the privacy benefits of on-premise deployment."
},
{
"venue": "EMNLP",
"title": "Multilingual Datasets for Custom Input Extraction and Explanation Requests Parsing in Conversational XAI Systems",
"authors": [
"Qianli Wang",
"Tatiana Anikina",
"Nils Feldhus",
"Simon Ostermann",
"F. Splitt",
"J.-L. F. Li",
"Yoana Tsoneva",
"Sebastian Möller",
"Vera Schmitt"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.29.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.29",
"abstract": "Qianli Wang, Tatiana Anikina, Nils Feldhus, Simon Ostermann, Fedor Splitt, Jiaao Li, Yoana Tsoneva, Sebastian Möller, Vera Schmitt. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "Is Translation All You Need? A Study on Solving Multilingual Tasks with Large Language Models",
"authors": [
"Chaoqun Liu",
"Wenxuan Zhang",
"Yiran Zhao",
"Anh Tuan Luu",
"Lidong Bing"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.naacl-long.485.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.naacl-long.485",
"abstract": "Chaoqun Liu, Wenxuan Zhang, Yiran Zhao, Anh Tuan Luu, Lidong Bing. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025."
},
{
"venue": "EMNLP",
"title": "GUARD: Glocal Uncertainty-Aware Robust Decoding for Effective and Efficient Open-Ended Text Generation",
"authors": [
"Y. Ding",
"Esteban Garces Arias",
"Meimingwei Li",
"Julian Rodemann",
"Matthias Aßenmacher",
"D. Chen",
"Gaojuan Fan",
"Christian Heumann",
"Chongsheng Zhang"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.380.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.380",
"abstract": "Yuanhao Ding, Esteban Garces Arias, Meimingwei Li, Julian Rodemann, Matthias Aßenmacher, Danlu Chen, Gaojuan Fan, Christian Heumann, Chongsheng Zhang. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "On explaining recommendations with Large Language Models: a review",
"authors": [
"Alan Said"
],
"year": 2025,
"pdf_url": "https://www.frontiersin.org/journals/big-data/articles/10.3389/fdata.2024.1505284/pdf",
"source": "openalex",
"doi": "https://doi.org/10.3389/fdata.2024.1505284",
"abstract": "The rise of Large Language Models (LLMs), such as LLaMA and ChatGPT, has opened new opportunities for enhancing recommender systems through improved explainability. This paper provides a systematic literature review focused on leveraging LLMs to generate explanations for recommendations-a critical aspect for fostering transparency and user trust. We conducted a comprehensive search within the ACM Guide to Computing Literature, covering publications from the launch of ChatGPT (November 2022) to the present (November 2024). Our search yielded 232 articles, but after applying inclusion criteria, only six were identified as directly addressing the use of LLMs in explaining recommendations. This scarcity highlights that, despite the rise of LLMs, their application in explainable recommender systems is still in an early stage. We analyze these select studies to understand current methodologies, identify challenges, and suggest directions for future research. Our findings underscore the potential of LLMs improving explanations of recommender systems and encourage the development of more transparent and user-centric recommendation explanation solutions."
},
{
"venue": "EMNLP",
"title": "A Survey of Large Language Models in Psychotherapy: Current Landscape and Future Directions",
"authors": [
"Hongbin Na",
"Yining Hua",
"Zimu Wang",
"Tao Shen",
"Beibei Yu",
"Lilin Wang",
"Wei Wang",
"John Torous",
"Ling Chen"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-acl.385.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-acl.385",
"abstract": "Mental health is increasingly critical in contemporary healthcare, with psychotherapy demanding dynamic, context-sensitive interactions that traditional NLP methods struggle to capture.Large Language Models (LLMs) offer significant potential for addressing this gap due to their ability to handle extensive context and multi-turn reasoning.This review introduces a conceptual taxonomy dividing psychotherapy into interconnected stages-assessment, diagnosis, and treatment-to systematically examine LLM advancements and challenges.Our comprehensive analysis reveals imbalances in current research, such as a focus on common disorders, linguistic biases, fragmented methods, and limited theoretical integration.We identify critical challenges including capturing dynamic symptom fluctuations, overcoming linguistic and cultural biases, and ensuring diagnostic reliability.Highlighting future directions, we advocate for continuous multi-stage modeling, real-time adaptive systems grounded in psychological theory, and diversified research covering broader mental disorders and therapeutic approaches, aiming toward more holistic and clinically integrated psychotherapy LLMs systems."
},
{
"venue": "EMNLP",
"title": "XCOT: Cross-lingual Instruction Tuning for Cross-lingual Chain-of-Thought Reasoning",
"authors": [
"Linzheng Chai",
"Jian Yang",
"Tao Sun",
"Hongcheng Guo",
"Jiaheng Liu",
"Bing Wang",
"Xinnian Liang",
"Jiaqi Bai",
"Tongliang Li",
"Qiyao Peng",
"Zhoujun Li"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/34524/36679",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i22.34524",
"abstract": "Chain-of-thought (CoT) has emerged as a powerful technique to elicit reasoning in large language models and improve a variety of downstream tasks. CoT mainly demonstrates excellent performance in English, but its usage in low-resource languages is constrained due to poor language generalization. To bridge the gap among different languages, we propose a cross-lingual instruction fine-tuning framework (xCoT) to transfer knowledge from high-resource languages to low-resource languages. Specifically, the multilingual instruction training data (xCoT-Instruct) is created to encourage the semantic alignment of multiple languages. We introduce cross-lingual in-context few-shot learning (xICL) to accelerate multilingual agreement in instruction tuning, where some fragments of source languages in examples are randomly substituted by their counterpart translations of target languages. During multilingual instruction tuning, we adopt the randomly online CoT strategy to enhance the multilingual reasoning ability of the large language model by first translating the query to another language and then answering in English. To further facilitate the language transfer, we leverage the high-resource CoT to supervise the training of low-resource languages with cross-lingual distillation. Experimental results demonstrate the superior performance of xCoT in reducing the gap among different languages, highlighting its potential to reduce the cross-lingual gap."
},
{
"venue": "EMNLP",
"title": "DeepNote: Note-Centric Deep Retrieval-Augmented Generation",
"authors": [
"Ruobing Wang",
"Qi Zhao",
"Yukun Yan",
"Daren Zha",
"Yuxuan Chen",
"Yu Shi",
"Zhenghao Liu",
"Yixuan Wang",
"Shuo Wang",
"Han Xu",
"Zhiyuan Liu",
"Maosong Sun"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.1073.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.1073",
"abstract": "Ruobing Wang, Qingfei Zhao, Yukun Yan, Daren Zha, Yuxuan Chen, Shi Yu, Zhenghao Liu, Yixuan Wang, Shuo Wang, Xu Han, Zhiyuan Liu, Maosong Sun. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "LoCal: Logical and Causal Fact-Checking with LLM-Based Multi-Agents",
"authors": [
"Jiatong Ma",
"Linmei Hu",
"Rang Li",
"Wenbo Fu"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714748",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714748",
"abstract": "With the development of social media, people are exposed to a vast amount of unverified information, making fact-checking particularly important. Existing fact-checking methods primarily encourage breaking down claims into more easily solvable sub-tasks, and deriving final answers through reasoning with external evidence. However, these models face logical issues regarding whether and how the sub-tasks can logically be combined to form the original claims, and encounter causal errors in the reasoning process due to insufficient evidence or hallucinations from LLMs. In addition, they often suffer from a lack of interpretability. In this paper, we propose Logical and Causal fact-checking (LoCal), a novel fact-checking framework based on multiple LLM-based agents. The usage of multi-agent systems is due to their increasingly demonstrated ability to perform complex tasks in a manner similar to humans. LoCal primarily consists of a decomposing agent, multiple reasoning agents, and two evaluating agents. Specifically, the decomposing agent first utilizes the in-context learning ability of LLMs to break down complex claims into simpler sub-tasks, including fact verification tasks and question answering tasks. Afterwards, two types of reasoning agents are respectively utilized to retrieve external knowledge to address the fact verification tasks that require comparative analysis skills, and the question answering tasks that necessitate the ability of information extraction from evidence. We then combine the sub-tasks and their corresponding responses to generate a solution for evaluation. In order to enhance logical and causal consistency, two evaluating agents are respectively employed to examine whether the generated solution is logically equivalent to the original claim and determine whether the solution still holds when challenged by the counterfactual label. The evaluating agents provide confidence degrees for the solutions based on the evaluation results and iteratively correct the logical and causal errors in the reasoning process. We evaluate LoCal on two challenging datasets, and the results show that LoCal significantly outperforms all the baseline models across different settings of evidence availability. In addition, LoCal offers better interpretability by providing a structured solution along with detailed evaluating processes. We believe LoCal will provide valuable insights for future misinformation detection."
},
{
"venue": "EMNLP",
"title": "Assessing and Understanding Creativity in Large Language Models",
"authors": [
"Yunpu Zhao",
"Rui Zhang",
"Wenyi Li",
"Ling Li"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s11633-025-1546-4.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s11633-025-1546-4",
"abstract": "Abstract In the field of natural language processing, the rapid development of large language model (LLM) has attracted increasing attention. LLMs have shown a high level of creativity in various tasks, but the methods for assessing such creativity are inadequate. Assessment of LLM creativity needs to consider differences from humans, requiring multiple dimensional measurement while balancing accuracy and efficiency. This paper aims to establish an efficient framework for assessing the level of creativity in LLMs. By adapting the modified Torrance tests of creative thinking, the research evaluates the creative performance of various LLMs across 7 tasks, emphasizing 4 criteria including fluency, flexibility, originality, and elaboration. In this context, we develop a comprehensive dataset of 700 questions for testing and an LLM-based evaluation method. In addition, this study presents a novel analysis of LLMs’ responses to diverse prompts and role-play situations. We found that the creativity of LLMs primarily falls short in originality, while excelling in elaboration. In addition, the use of prompts and role-play settings of the model significantly influence creativity. Additionally, the experimental results also indicate that collaboration among multiple LLMs can enhance originality. Notably, our findings reveal a consensus between human evaluations and LLMs regarding the personality traits that influence creativity. The findings underscore the significant impact of LLM design on creativity and bridge artificial intelligence and human creativity, offering insights into LLMs’ creativity and potential applications."
},
{
"venue": "EMNLP",
"title": "MFTCXplain: A Multilingual Benchmark Dataset for Evaluating the Moral Reasoning of LLMs through Multi-hop Hate Speech Explanation",
"authors": [
"Jackson Trager",
"Francielle Vargas",
"Diego Alves",
"M. Guida",
"Mikel K. Ngueajio",
"Ameeta Agrawal",
"Yalda Daryani",
"Farzan Karimi Malekabadi",
"Flor Miriam Plaza-del-Arco"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.851.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.851",
"abstract": "Jackson Trager, Francielle Vargas, Diego Alves, Matteo Guida, Mikel K. Ngueajio, Ameeta Agrawal, Yalda Daryani, Farzan Karimi Malekabadi, Flor Miriam Plaza-del-Arco. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "LLM-Prop: predicting the properties of crystalline materials using large language models",
"authors": [
"Andre Niyongabo Rubungo",
"Craig B. Arnold",
"Barry P. Rand",
"Adji Bousso Dieng"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41524-025-01536-2.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41524-025-01536-2",
"abstract": "Abstract The prediction of crystal properties plays a crucial role in materials science and applications. Current methods for predicting crystal properties focus on modeling crystal structures using graph neural networks (GNNs). However, accurately modeling the complex interactions between atoms and molecules within a crystal remains a challenge. Surprisingly, predicting crystal properties from crystal text descriptions is understudied, despite the rich information and expressiveness that text data offer. In this paper, we develop and make public a benchmark dataset (TextEdge) that contains crystal text descriptions with their properties. We then propose LLM-Prop, a method that leverages the general-purpose learning capabilities of large language models (LLMs) to predict properties of crystals from their text descriptions. LLM-Prop outperforms the current state-of-the-art GNN-based methods by approximately 8% on predicting band gap, 3% on classifying whether the band gap is direct or indirect, and 65% on predicting unit cell volume, and yields comparable performance on predicting formation energy per atom, energy per atom, and energy above hull. LLM-Prop also outperforms the fine-tuned MatBERT, a domain-specific pre-trained BERT model, despite having 3 times fewer parameters. We further fine-tune the LLM-Prop model directly on CIF files and condensed structure information generated by Robocrystallographer and found that LLM-Prop fine-tuned on text descriptions provides a better performance on average. Our empirical results highlight the importance of having a natural language input to LLMs to accurately predict crystal properties and the current inability of GNNs to capture information pertaining to space group symmetry and Wyckoff sites for accurate crystal property prediction."
},
{
"venue": "EMNLP",
"title": "HtmlRAG: HTML is Better Than Plain Text for Modeling Retrieved Knowledge in RAG Systems",
"authors": [
"Jiejun Tan",
"Zhicheng Dou",
"Wen Wang",
"M. W. Wang",
"Weipeng Chen",
"Ji-Rong Wen"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714546",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714546",
"abstract": "Retrieval-Augmented Generation (RAG) has been shown to improve knowledge capabilities and alleviate the hallucination problem of LLMs. The Web is a major source of external knowledge used in RAG systems, and many commercial RAG systems have used Web search engines as their major retrieval systems. Typically, such RAG systems retrieve search results, download HTML sources of the results, and then extract plain texts from the HTML sources. Plain text documents or chunks are fed into the LLMs to augment the generation. However, much of the structural and semantic information inherent in HTML, such as headings and table structures, is lost during this plain-text-based RAG process. To alleviate this problem, we propose HtmlRAG, which uses HTML instead of plain text as the format of retrieved knowledge in RAG. We believe HTML is better than plain text in modeling knowledge in external documents, and most LLMs possess robust capacities to understand HTML. However, utilizing HTML presents new challenges. HTML contains additional content such as tags, JavaScript, and CSS specifications, which bring extra input tokens and noise to the RAG system. To address this issue, we propose HTML cleaning, compression, and a two-step block-tree-based pruning strategy, to shorten the HTML while minimizing the loss of information. Experiments on six QA datasets confirm the superiority of using HTML in RAG systems. Our code and datasets are available at https://github.com/plageon/HtmlRAG."
},
{
"venue": "EMNLP",
"title": "Prompting large language models with knowledge graphs for question answering involving long-tail facts",
"authors": [
"Wenyu Huang",
"Guancheng Zhou",
"Mirella Lapata",
"Pavlos Vougiouklis",
"Sébastien Montella",
"Jeff Z. Pan"
],
"year": 2025,
"pdf_url": "https://www.research.ed.ac.uk/en/publications/34c2edb2-1fa3-4e5b-9609-3f68d1545203",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.knosys.2025.113648",
"abstract": ""
},
{
"venue": "EMNLP",
"title": "A Novel Playbook for Pragmatic Trial Operations to Monitor and Evaluate Ambient Artificial Intelligence in Clinical Practice",
"authors": [
"Majid Afshar",
"Felice Resnik",
"Mary Ryan",
"Josie Hintzke",
"Kayla K. Lemmon",
"Anne Gravel Sullivan",
"Tina Shah",
"Anthony Stordalen",
"Michael Oberst",
"Jason Dambach",
"Leigh A. Mrotek",
"Mariah A. Quinn",
"Kirsten Abramson",
"Peter Kleinschmidt",
"Tom Brazelton",
"Heidi Twedt",
"David Kunstman",
"Graham Wills",
"John Long",
"Brian W. Patterson",
"Frank Liao",
"Stacy Rasmussen",
"Elizabeth S. Burnside",
"Cherodeep Goswami",
"Joel Gordon"
],
"year": 2025,
"pdf_url": "https://ai.nejm.org/doi/pdf/10.1056/AIdbp2401267",
"source": "openalex",
"doi": "https://doi.org/10.1056/aidbp2401267",
"abstract": "BACKGROUND: Ambient artificial intelligence (AI) offers the potential to reduce documentation burden and improve efficiency through clinical note generation. Widespread adoption, however, remains limited due to challenges in electronic health record (EHR) integration, coding compliance, and real-world evaluation. This study introduces a framework and protocols to design, monitor, and deploy ambient AI within routine care. METHODS: , Tenth Revision (ICD-10) compliance were performed using an internally developed large language model (LLM), the validity of which was assessed through correlation with certified professional coders. RESULTS: Ambient AI utilization, measured as the proportion of eligible clinical notes completed using the system, had a weighted median of 65.4% (interquartile range, 50.6 to 84.0%). Iterative improvement cycles targeted task-specific adoption. A brief workflow issue related to a note template change initially reduced ICD-10 documentation accuracy from 79% (95% confidence interval [CI], 72 to 86%) to 35% (95% CI, 28 to 42%); accuracy returned to baseline after note template redesign and user training. The internally developed LLM coder achieved a strong correlation with professional coders (Pearson's r=0.97). The trial enrolled 66 providers across eight specialties, powered at 90% for the primary outcome of provider well-being. CONCLUSIONS: We provide a publicly available framework and protocols to help safely implement ambient AI in health care. Innovations include an embedded pragmatic trial design, human factors engineering, compliance-driven feedback loops, and real-time monitoring to support deployment, ensuring fidelity before initiation of the clinical trial. (Funded by the University of Wisconsin Hospital and Clinics and the National Institutes of Health Clinical and Translational Science Award; NIH/ NCATS UL1TR002737; ClinicalTrials.gov number, NCT06517082.)."
},
{
"venue": "EMNLP",
"title": "Tracing Multilingual Factual Knowledge Acquisition in Pretraining",
"authors": [
"Yihong Liu",
"Mingyang Wang",
"Amir Hossein Kargaran",
"Felicia Körner",
"Ercong Nie",
"Barbara Plank",
"François Yvon",
"Hinrich Schuetze"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.113.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.113",
"abstract": "Yihong Liu, Mingyang Wang, Amir Hossein Kargaran, Felicia Körner, Ercong Nie, Barbara Plank, François Yvon, Hinrich Schuetze. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "The foundational capabilities of large language models in predicting postoperative risks using clinical notes",
"authors": [
"Charles Alba",
"Bing Xue",
"Joanna Abraham",
"Thomas Kannampallil",
"Chenyang Lu"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41746-025-01489-2.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41746-025-01489-2",
"abstract": "Clinical notes recorded during a patient's perioperative journey holds immense informational value. Advances in large language models (LLMs) offer opportunities for bridging this gap. Using 84,875 preoperative notes and its associated surgical cases from 2018 to 2021, we examine the performance of LLMs in predicting six postoperative risks using various fine-tuning strategies. Pretrained LLMs outperformed traditional word embeddings by an absolute AUROC of 38.3% and AUPRC of 33.2%. Self-supervised fine-tuning further improved performance by 3.2% and 1.5%. Incorporating labels into training further increased AUROC by 1.8% and AUPRC by 2%. The highest performance was achieved with a unified foundation model, with improvements of 3.6% for AUROC and 2.6% for AUPRC compared to self-supervision, highlighting the foundational capabilities of LLMs in predicting postoperative risks, which could be potentially beneficial when deployed for perioperative care."
},
{
"venue": "EMNLP",
"title": "Theory and Toolkits for User Simulation in the Era of Generative AI: User Modeling, Synthetic Data Generation, and System Evaluation",
"authors": [
"Krisztian Balog",
"Nolwenn Bernard",
"Saber Zerhoudi",
"ChengXiang Zhai"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3731697",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3731697",
"abstract": "Interactive AI systems, including search engines, recommender systems, conversational agents, and generative AI applications, are increasingly central to user experiences. However, rigorously evaluating their performance, training them effectively with interaction data, and modeling user behavior for personalization remain significant challenges, often difficult to address reproducibly and at scale. User simulation, which employs intelligent agents to mimic human interaction patterns, offers a powerful and versatile methodology to tackle these interconnected issues. This half-day tutorial provides a comprehensive overview of modern user simulation techniques for interactive AI systems. We will explore the theoretical foundations and practical applications of simulation for system evaluation, algorithm training, and user modeling, emphasizing the crucial connections between these uses. The tutorial covers key simulation methodologies, with a particular focus on recent advancements leveraging large language models, discussing both the opportunities they present and the open challenges they entail. Crucially, we will also provide practical guidance, highlighting relevant toolkits, libraries, and datasets available to researchers and practitioners."
},
{
"venue": "EMNLP",
"title": "Efficient GPT-4V level multimodal large language model for deployment on edge devices",
"authors": [
"Yuan Yao",
"Tianyu Yu",
"Ao Zhang",
"Chongyi Wang",
"Junbo Cui",
"Hongji Zhu",
"Tianchi Cai",
"Chi Chen",
"Haoyu Li",
"Weilin Zhao",
"Zhihui He",
"Qianyu Chen",
"R. Zhou",
"Zhensheng Zou",
"Haoye Zhang",
"Shengding Hu",
"Zhi Zheng",
"Jie Zhou",
"Jie Cai",
"Han Xu",
"Guoyang Zeng",
"Dahai Li",
"Zhiyuan Liu",
"Maosong Sun"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41467-025-61040-5.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41467-025-61040-5",
"abstract": "Multimodal large language models have revolutionized AI research and industry, paving the way toward the next milestone. However, their large sizes and high computational costs restrict deployment to cloud servers, limiting use in mobile, offline, energy-sensitive, or privacy-critical scenarios. We present MiniCPM-V, efficient models for edge devices that integrate advancements in architecture, training, and data. The 8B model outperforms GPT-4V, Gemini Pro, and Claude 3 across 11 public benchmarks, processes high-resolution images at any aspect ratio, achieves robust optical character recognition, exhibits low hallucination rates, and supports over 30 languages while running efficiently on mobile phones. This progress reflects a broader trend: The sizes for high-performing models are rapidly decreasing alongside growing edge computation capacity, enabling advanced multimodal models to operate locally on consumer hardware. Such developments unlock applications across diverse real-world scenarios, from enhanced mobile AI to privacy-preserving solutions, marking a critical step toward democratizing powerful multimodal intelligence."
},
{
"venue": "EMNLP",
"title": "prompt4vis: prompting large language models with example mining for tabular data visualization",
"authors": [
"Shuaimin Li",
"Xuanang Chen",
"Yuanfeng Song",
"Yunze Song",
"Chen Zhang",
"Fei Hao",
"Lei Chen"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s00778-025-00912-0.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s00778-025-00912-0",
"abstract": "Abstract We are currently in the epoch of Large Language Models (LLMs), which have transformed numerous technological domains within the database community. In this paper, we examine the application of LLMs in text-to-visualization (text-to-vis). The advancement of natural language processing technologies has made natural language interfaces more accessible and intuitive for visualizing tabular data. However, despite utilizing advanced neural network architectures, current methods such as Seq2Vis, ncNet, and RGVisNet for transforming natural language queries into DV commands still underperform, indicating significant room for improvement. In this paper, we introduce Prompt4Vis , a novel framework that leverages LLMs and In-context learning to enhance the generation of data visualizations from natural language. Given that In-context learning’s effectiveness is highly dependent on the selection of examples, it is critical to optimize this aspect. Additionally, encoding the full database schema of a query is not only costly but can also lead to inaccuracies. This framework includes two main components: (1) an example mining module that identifies highly effective examples to enhance In-context learning capabilities for text-to-vis applications, and (2) a schema filtering module designed to streamline database schemas. Comprehensive testing on the NVBench dataset has shown that Prompt4Vis significantly outperforms the current state-of-the-art model, RGVisNet, by approximately 35.9% on development sets and 71.3% on test sets. To the best of our knowledge, Prompt4Vis is the first framework to incorporate In-context learning for enhancing text-to-vis, marking a pioneering step in the domain."
},
{
"venue": "EMNLP",
"title": "KG-Agent: An Efficient Autonomous Agent Framework for Complex Reasoning over Knowledge Graph",
"authors": [
"Jinhao Jiang",
"Kun Zhou",
"Xin Zhao",
"Yang Song",
"Chen Zhu",
"Hengshu Zhu",
"Ji-Rong Wen"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.acl-long.468.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.acl-long.468",
"abstract": "In this paper, we aim to improve the reasoning ability of large language models (LLMs) over knowledge graphs (KGs) to answer complex questions.Inspired by existing methods that design the interaction strategy between LLMs and KG, we propose an autonomous LLM-based agent framework, called KG-Agent, which enables a small LLM to actively make decisions until finishing the reasoning process over KGs.In KG-Agent, we integrate the LLM, multifunctional toolbox, KG-based executor, and knowledge memory, and develop an iteration mechanism that autonomously selects the tool and then updates the memory for reasoning over KG.To guarantee the effectiveness, we leverage program language to formulate the multi-hop reasoning process over the KG and synthesize a code-based instruction dataset to fine-tune the base LLM.Extensive experiments demonstrate that only using 10K samples for tuning LLaMA2-7B can outperform competitive methods using larger LLMs or more data, on both in-domain and out-domain datasets.Our code and data will be publicly released."
},
{
"venue": "EMNLP",
"title": "Show Me the Work: Fact-Checkers' Requirements for Explainable Automated Fact-Checking",
"authors": [
"Greta Warren",
"Irina Shklovski",
"Isabelle Augenstein"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3706598.3713277",
"source": "openalex",
"doi": "https://doi.org/10.1145/3706598.3713277",
"abstract": "The pervasiveness of large language models and generative AI in online media has amplified the need for effective automated fact-checking to assist fact-checkers in tackling the increasing volume and sophistication of misinformation. The complex nature of fact-checking demands that automated fact-checking systems provide explanations that enable fact-checkers to scrutinise their outputs. However, it is unclear how these explanations should align with the decision-making and reasoning processes of fact-checkers to be effectively integrated into their workflows. Through semi-structured interviews with fact-checking professionals, we bridge this gap by: (i) providing an account of how fact-checkers assess evidence, make decisions, and explain their processes; (ii) examining how fact-checkers use automated tools in practice; and (iii) identifying fact-checker explanation requirements for automated fact-checking tools. The findings show unmet explanation needs and identify important criteria for replicable fact-checking explanations that trace the model's reasoning path, reference specific evidence, and highlight uncertainty and information gaps."
},
{
"venue": "EMNLP",
"title": "X-CoT: Explainable Text-to-Video Retrieval via LLM-based Chain-of-Thought Reasoning",
"authors": [
"Pulakurthi, Prasanna Reddy",
"Wang, Jiamian",
"Rabbani, Majid",
"Dianat, Sohail",
"Rao, Raghuveer",
"Tao, Zhiqiang"
],
"year": 2025,
"pdf_url": "https://doi.org/10.5281/zenodo.17613043",
"source": "openalex",
"doi": "https://doi.org/10.5281/zenodo.17613043",
"abstract": "Official implementation of \"X-CoT: Explainable Text-to-Video Retrieval via LLM-based Chain-of-Thought Reasoning (EMNLP 2025)\""
},
{
"venue": "EMNLP",
"title": "Optimised knowledge distillation for efficient social media emotion recognition using DistilBERT and ALBERT",
"authors": [
"Muhammad Hussain",
"Caikou Chen",
"Muzammil Hussain",
"Muhammad Tuoqeer Anwar",
"Mohammed Abaker",
"Abdelzahir Abdelmaboud",
"Iqra Yamin"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-16001-9.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-16001-9",
"abstract": "Accurate emotion recognition in social media text is critical for applications such as sentiment analysis, mental health monitoring, and human-computer interaction. However, existing approaches face challenges like computational complexity and class imbalance, limiting their deployment in resource-constrained environments. While transformer-based models achieve state-of-the-art performance, their size and latency hinder real-time applications. To address these issues, we propose a novel knowledge distillation framework that transfers knowledge from a fine-tuned BERT-base teacher model to lightweight DistilBERT and ALBERT student models, optimised for efficient emotion recognition. Our approach integrates a hybrid loss function combining focal loss and Kullback-Leibler (KL) divergence to enhance minority class recognition, attention-head alignment for effective contextual knowledge transfer, and semantic-preserving data augmentation to mitigate class imbalance. Experiments on two datasets, Twitter Emotions 416 K samples, six classes, and Social Media Emotion 75 K samples, five classes, show that our distilled models achieve near-teacher performance 97.35% and 73.86% accuracy, respectively. with only a < 1% and < 6% accuracy drop, while reducing model size by 40% and inference latency by 3.2×. Notably, our method significantly improves F1-scores for minority classes. Our work sets a new state-of-the-art in efficient emotion recognition, enabling practical deployment in edge computing and mobile applications."
},
{
"venue": "EMNLP",
"title": "RG-VQA: Leveraging Retriever-Generator Pipelines for Knowledge Intensive Visual Question Answering",
"authors": [
"Settaluri Lakshmi Sravanthi",
"Pulkit Agarwal",
"Debjyoti Mondal",
"Rituraj Singh",
"Subhadarshi Panda",
"Ankit Mishra",
"K R Pradeep",
"K. Srihari",
"Godawari Sudhakar Rao",
"Pushpak Bhattacharyya"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.1306.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.1306",
"abstract": "Settaluri Lakshmi Sravanthi, Pulkit Agarwal, Debjyoti Mondal, Rituraj Singh, Subhadarshi Panda, Ankit Mishra, Kiran Pradeep, Srihari K B, Godawari Sudhakar Rao, Pushpak Bhattacharyya. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "Bridging the Creativity Understanding Gap: Small-Scale Human Alignment Enables Expert-Level Humor Ranking in LLMs",
"authors": [
"Kuan Lok Zhou",
"Jiayi Chen",
"Siddharth Suresh",
"Reuben Narad",
"Timothy T. Rogers",
"Lalit Jain",
"Robert D. Nowak",
"Bob Mankoff",
"Jifan Zhang"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.884.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.884",
"abstract": "Kuan Lok Zhou, Jiayi Chen, Siddharth Suresh, Reuben Narad, Timothy T. Rogers, Lalit K Jain, Robert D Nowak, Bob Mankoff, Jifan Zhang. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "Improving Retrieval Augmented Language Model with Self-Reasoning",
"authors": [
"Yuan Xia",
"Jingbo Zhou",
"Zhenhui Shi",
"Jun Chen",
"Haifeng Huang"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/34743/36898",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i24.34743",
"abstract": "The Retrieval-Augmented Language Model (RALM) has demonstrated remarkable performance on knowledge-intensive tasks by integrating external knowledge during inference, which mitigates the factual hallucinations inherited in large language models (LLMs). Despite these advancements, challenges persist in the implementation of RALMs, particularly in terms of reliability and traceability. Specifically, the irrelevant document retrieval may result in unhelpful responses or even deteriorate the performance of LLMs, while the lack of appropriate citations in outputs complicates efforts to verify the trustworthiness of the models. To this end, we propose a novel self-reasoning framework aimed at improving the reliability and traceability of RALMs, whose core idea is to leverage reasoning trajectories generated by the LLM itself. The framework involves constructing self-reasoning trajectories through three processes: a relevance-aware process, an evidence-aware selective process, and a trajectory analysis process. We evaluated our framework across four public datasets (two short-form QA datasets, one long-form QA dataset, and one fact verification dataset) to demonstrate its superiority. Our method can outperform existing state-of-the-art models and achieve performance comparable with GPT-4, using only 2,000 training samples."
},
{
"venue": "EMNLP",
"title": "Watermarking for Large Language Models: A Survey",
"authors": [
"Zhiguang Yang",
"Gejian Zhao",
"Hanzhou Wu"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2227-7390/13/9/1420/pdf?version=1745651907",
"source": "openalex",
"doi": "https://doi.org/10.3390/math13091420",
"abstract": "With the rapid advancement and widespread deployment of large language models (LLMs), concerns regarding content provenance, intellectual property protection, and security threats have become increasingly prominent. Watermarking techniques have emerged as a promising solution for embedding verifiable signals into model outputs, enabling attribution, authentication, and mitigation of unauthorized usage. Despite growing interest in watermarking LLMs, the field lacks a systematic review to consolidate existing research and assess the effectiveness of different techniques. Key challenges include the absence of a unified taxonomy and limited understanding of trade-offs between capacity, robustness, and imperceptibility in real-world scenarios. This paper addresses these gaps by providing a comprehensive survey of watermarking methods tailored to LLMs, structured around three core contributions: (1) We classify these methods as training-free and training-based approaches and detail their mechanisms, strengths, and limitations to establish a structured understanding of existing techniques. (2) We evaluate these techniques based on key criteria—including robustness, imperceptibility, and payload capacity—to identify their effectiveness and limitations, highlighting challenges in designing resilient and practical watermarking solutions. (3) We also discuss critical open challenges while outlining future research directions and practical considerations to drive innovation in watermarking for LLMs. By providing a structured synthesis, this work advances the development of secure and effective watermarking solutions for LLMs."
},
{
"venue": "EMNLP",
"title": "CLAPnq: Cohesive Long-form Answers from Passages in Natural Questions for RAG systems",
"authors": [
"Sara Rosenthal",
"Avirup Sil",
"Radu Florian",
"Salim Roukos"
],
"year": 2025,
"pdf_url": "https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00729/2499744/tacl_a_00729.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1162/tacl_a_00729",
"abstract": "Abstract Retrieval Augmented Generation (RAG) has become a popular application for large language models. It is preferable that successful RAG systems provide accurate answers that are supported by being grounded in a passage without any hallucinations. While considerable work is required for building a full RAG pipeline, being able to benchmark performance is also necessary. We present CLAPnq, a benchmark Long-form Question Answering dataset for the full RAG pipeline. CLAPnq includes long answers with grounded gold passages from Natural Questions (NQ) and a corpus to perform either retrieval, generation, or the full RAG pipeline. The CLAPnq answers are concise, 3x smaller than the full passage, and cohesive, meaning that the answer is composed fluently, often by integrating multiple pieces of the passage that are not contiguous. RAG models must adapt to these properties to be successful at CLAPnq. We present baseline experiments and analysis for CLAPnq that highlight areas where there is still significant room for improvement in grounded RAG. CLAPnq is publicly available at https://github.com/primeqa/clapnq."
},
{
"venue": "EMNLP",
"title": "Can AI support human grading? Examining machine attention and confidence in short answer scoring",
"authors": [
"Yuheng Li",
"Mladen Raković",
"Namrata Srivastava",
"Xinyu Li",
"Quanlong Guan",
"Dragan Gašević",
"Guanliang Chen"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.compedu.2025.105244",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.compedu.2025.105244",
"abstract": "Large language models built upon artificial intelligence (AI) hold great promises to innovate automatic short answer scoring (ASAS) - significantly alleviating educators’ workload in assessing student answers. However, ASAS systems on such basis have seen limited adoption in authentic teaching environments due to the models’ inability to explain the predictions they generate. To address this, we recruited 32 human graders to comparatively analyse the decision-making processes of human graders and AI-driven graders. Specifically, we exploited two types of data to holistically unveil the decision-making processes of human graders throughout grading, namely manual annotation of important words and gaze data of the human graders. The decision-making processes of AI-driven graders were revealed by important words extracted though eXplainable Artificial Intelligence technique and grading confidence reflected by the prediction probability distributions. We measured the alignment in their decision-making regarding their (i) estimated scoring difficulty, (ii) important text segments and (iii) crucial grammatical categories to enhance the transparency and trustworthiness of AI-driven graders. Subsequently, we conducted randomised control studies, presenting machine-extracted insights like important words and estimated scoring difficulty to scrutinise how they affected human grading. Our findings contribute new knowledge regarding the consistency between human and machine scoring and validates machine-extracted insights, such as important words and scoring difficulty, to be valuable in facilitating human grading, encouraging the adoption of ASAS systems and urging the potential collaboration between machine and human grading in pedagogical practices. However, we emphasised the significance of grasping question context and intricacy before leveraging such machine-extracted insights. • Machine and human graders share similar estimations of scoring difficulty. • Machine and human graders focus on a similar set of words to assess answer quality. • Insights drawn from machine grading hold great promises to facilitate human grading."
},
{
"venue": "EMNLP",
"title": "Towards evaluating and building versatile large language models for medicine",
"authors": [
"Chaoyi Wu",
"Pengcheng Qiu",
"Jinxin Liu",
"Hongfei Gu",
"Na Li",
"Ya Zhang",
"Yanfeng Wang",
"Weidi Xie"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41746-024-01390-4.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41746-024-01390-4",
"abstract": "In this study, we present MedS-Bench, a comprehensive benchmark to evaluate large language models (LLMs) in clinical contexts, MedS-Bench, spanning 11 high-level clinical tasks. We evaluate nine leading LLMs, e.g., MEDITRON, Llama 3, Mistral, GPT-4, Claude-3.5, etc. and found that most models struggle with these complex tasks. To address these limitations, we developed MedS-Ins, a large-scale instruction-tuning dataset for medicine. MedS-Ins comprises 58 medically oriented language corpora, totaling 5M instances with 19K instructions, across 122 tasks. To demonstrate the dataset's utility, we conducted a proof-of-concept experiment by performing instruction tuning on a lightweight, open-source medical language model. The resulting model, MMedIns-Llama 3, significantly outperformed existing models on various clinical tasks. To promote further advancements, we have made MedS-Ins fully accessible and invite the research community to contribute to its expansion. Additionally, we have launched a dynamic leaderboard for MedS-Bench, to track the development progress of medical LLMs."
},
{
"venue": "EMNLP",
"title": "Qwen3 Technical Report",
"authors": [
"An Yang",
"Anfeng Li",
"Baosong Yang",
"Beichen Zhang",
"Binyuan Hui",
"Bo Zheng",
"B. X. Yu",
"Chang Gao",
"C. Huang",
"Chenxu Lv",
"Chujie Zheng",
"Dayiheng Liu",
"Fan Zhou",
"Fei Huang",
"H Feng",
"Hao Ge",
"Haoran Wei",
"Lin Huan",
"Jialong Tang",
"Jian Yang",
"Jianhong Tu",
"Jianwei Zhang",
"Jianxin Yang",
"Jiaxi Yang",
"Jing Zhou",
"Jingren Zhou",
"Junyang Lin",
"Kai Dang",
"Keqin Bao",
"Kexin Yang",
"Le Yu",
"Deng, Lianghao",
"Mei Li",
"Mingfeng Xue",
"Mingze Li",
"Pei Zhang",
"Peng Wang",
"Qin Zhu",
"Rui Men",
"Ruize Gao",
"Shixuan Liu",
"Shuang Luo",
"Tianhao Li",
"Tianyi Tang",
"Wenbiao Yin",
"Xingzhang Ren",
"Xinyu Wang",
"Xinyu Zhang",
"Xuancheng Ren",
"Fan Yang",
"Su Yang",
"Yichang Zhang",
"Yinger Zhang",
"Yu Wan",
"Yuqiong Liu",
"Zekun Wang",
"Zeyu Cui",
"Zhenru Zhang",
"Zhipeng Zhou",
"Zihan Qiu"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2505.09388",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2505.09388",
"abstract": "In this work, we present Qwen3, the latest version of the Qwen model family. Qwen3 comprises a series of large language models (LLMs) designed to advance performance, efficiency, and multilingual capabilities. The Qwen3 series includes models of both dense and Mixture-of-Expert (MoE) architectures, with parameter scales ranging from 0.6 to 235 billion. A key innovation in Qwen3 is the integration of thinking mode (for complex, multi-step reasoning) and non-thinking mode (for rapid, context-driven responses) into a unified framework. This eliminates the need to switch between different models--such as chat-optimized models (e.g., GPT-4o) and dedicated reasoning models (e.g., QwQ-32B)--and enables dynamic mode switching based on user queries or chat templates. Meanwhile, Qwen3 introduces a thinking budget mechanism, allowing users to allocate computational resources adaptively during inference, thereby balancing latency and performance based on task complexity. Moreover, by leveraging the knowledge from the flagship models, we significantly reduce the computational resources required to build smaller-scale models, while ensuring their highly competitive performance. Empirical evaluations demonstrate that Qwen3 achieves state-of-the-art results across diverse benchmarks, including tasks in code generation, mathematical reasoning, agent tasks, etc., competitive against larger MoE models and proprietary models. Compared to its predecessor Qwen2.5, Qwen3 expands multilingual support from 29 to 119 languages and dialects, enhancing global accessibility through improved cross-lingual understanding and generation capabilities. To facilitate reproducibility and community-driven research and development, all Qwen3 models are publicly accessible under Apache 2.0."
},
{
"venue": "EMNLP",
"title": "Leveraging High-Resource English Corpora for Cross-lingual Domain Adaptation in Low-Resource Japanese Medicine via Continued Pre-training",
"authors": [
"Kazuma Kobayashi",
"Zhen Wan",
"Fei Cheng",
"Yuma Tsuta",
"Xin Zhao",
"Junfeng Jiang",
"Jiahao Huang",
"Zhiyi Huang",
"Yusuke Oda",
"Rio Yokota",
"Yuki Arase",
"Daisuke Kawahara",
"Akiko Aizawa",
"Sadao Kurohashi"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.615.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.615",
"abstract": "Kazuma Kobayashi, Zhen Wan, Fei Cheng, Yuma Tsuta, Xin Zhao, Junfeng Jiang, Jiahao Huang, Zhiyi Huang, Yusuke Oda, Rio Yokota, Yuki Arase, Daisuke Kawahara, Akiko Aizawa, Sadao Kurohashi. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025."
},
{
"venue": "EMNLP",
"title": "Amplifying Minority Voices: AI-Mediated Devil's Advocate System for Inclusive Group Decision-Making",
"authors": [
"Soohwan Lee",
"Mingyu Kim",
"Seoyeong Hwang",
"Da-jung Kim",
"Kyungho Lee"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2502.06251",
"source": "openalex",
"doi": "https://doi.org/10.1145/3708557.3716334",
"abstract": "Group decision-making often benefits from diverse perspectives, yet power imbalances and social influence can stifle minority opinions and compromise outcomes. This prequel introduces an AI-mediated communication system that leverages the Large Language Model to serve as a devil's advocate, representing underrepresented viewpoints without exposing minority members' identities. Rooted in persuasive communication strategies and anonymity, the system aims to improve psychological safety and foster more inclusive decision-making. Our multi-agent architecture, which consists of a summary agent, conversation agent, AI duplicate checker, and paraphrase agent, encourages the group's critical thinking while reducing repetitive outputs. We acknowledge that reliance on text-based communication and fixed intervention timings may limit adaptability, indicating pathways for refinement. By focusing on the representation of minority viewpoints anonymously in power-imbalanced settings, this approach highlights how AI-driven methods can evolve to support more divergent and inclusive group decision-making."
},
{
"venue": "EMNLP",
"title": "Multimodal generative AI for interpreting 3D medical images and videos",
"authors": [
"Jung-Oh Lee",
"Hong-Yu Zhou",
"Tyler M. Berzin",
"Daniel K. Sodickson",
"Pranav Rajpurkar"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41746-025-01649-4.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41746-025-01649-4",
"abstract": "This perspective proposes adapting video-text generative AI to 3D medical imaging (CT/MRI) and medical videos (endoscopy/laparoscopy) by treating 3D images as videos. The approach leverages modern video models to analyze multiple sequences simultaneously and provide real-time AI assistance during procedures. The paper examines medical imaging's unique characteristics (synergistic information, metadata, and world model), outlines applications in automated reporting, case retrieval, and education, and addresses challenges of limited datasets, benchmarks, and specialized training."
},
{
"venue": "EMNLP",
"title": "Query Performance Prediction Using Relevance Judgments Generated by Large Language Models",
"authors": [
"Chuan Meng",
"Negar Arabzadeh",
"Arian Askari",
"Mohammad Aliannejadi",
"Maarten de Rijke"
],
"year": 2025,
"pdf_url": "https://pure.uva.nl/ws/files/308953815/Query_Performance_Prediction_Using_Relevance_Judgments_Generated_by_Large_Language_Models.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1145/3736402",
"abstract": "Query performance prediction (QPP) aims to estimate the retrieval quality of a search system for a query without human relevance judgments. Previous QPP methods typically return a single scalar value and do not require the predicted values to approximate a specific information retrieval (IR) evaluation measure, leading to certain drawbacks: (i) a single scalar is insufficient to accurately represent different IR evaluation measures, especially when metrics do not highly correlate, and (ii) a single scalar limits the interpretability of QPP methods because solely using a scalar is insufficient to explain QPP results. To address these issues, we propose a QPP framework using automatically gen erated re levance judgments (QPP-GenRE), which decomposes QPP into independent subtasks of predicting the relevance of each item in a ranked list to a given query. This allows us to predict any IR evaluation measure using the generated relevance judgments as pseudo-labels. This also allows us to interpret predicted IR evaluation measures, and identify, track, and rectify errors in generated relevance judgments to improve QPP quality. We predict an item’s relevance by using open source large language models (LLMs) to ensure scientific reproducibility. We face two main challenges: (i) excessive computational costs of judging an entire corpus for predicting a metric considering recall, and (ii) limited performance in prompting open source LLMs in a zero-/few-shot manner. To solve the challenges, we devise an approximation strategy to predict an IR measure considering recall and propose to fine-tune open source LLMs using human-labeled relevance judgments. Experiments on the TREC 2019–2022 deep learning tracks and CAsT-19–20 datasets show that QPP-GenRE achieves state-of-the-art QPP quality for both lexical and neural rankers."
},
{
"venue": "CHIIR",
"title": "NeuroPhysIIR: International Workshop on NeuroPhysiological Approaches for Interactive Information Retrieval",
"authors": [
"Jacek Gwizdka",
"Javed Mostafa",
"Min Zhang",
"Kaixin Ji",
"Yashar Moshfeghi",
"Tuukka Ruotsalo",
"Damiano Spina"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3698204.3716481",
"source": "openalex",
"doi": "https://doi.org/10.1145/3698204.3716481",
"abstract": "The International Workshop on NeuroPhysiological Approaches for Interactive Information Retrieval (NeuroPhysIIR'25) aims to bringing together researchers from information science, humancomputer interaction, cognitive neuroscience, and related fields, to foster cross-disciplinary collaboration and accelerate progress in neurophysiologically-informed IIR research.As the third edition following successful workshops at SIGIR'15 [5] and CHIIR '17 [6], we anticipate that the interactive nature of this workshop will not only raise awareness but also lower the entry barriers for engaging with this exciting research area within the wider IIR community.Workshop website: https://neurophysiir.github.io/chiir2025/."
},
{
"venue": "CHIIR",
"title": "Theory and Toolkits for User Simulation in the Era of Generative AI: User Modeling, Synthetic Data Generation, and System Evaluation",
"authors": [
"Krisztian Balog",
"Nolwenn Bernard",
"Saber Zerhoudi",
"ChengXiang Zhai"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3731697",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3731697",
"abstract": "Interactive AI systems, including search engines, recommender systems, conversational agents, and generative AI applications, are increasingly central to user experiences. However, rigorously evaluating their performance, training them effectively with interaction data, and modeling user behavior for personalization remain significant challenges, often difficult to address reproducibly and at scale. User simulation, which employs intelligent agents to mimic human interaction patterns, offers a powerful and versatile methodology to tackle these interconnected issues. This half-day tutorial provides a comprehensive overview of modern user simulation techniques for interactive AI systems. We will explore the theoretical foundations and practical applications of simulation for system evaluation, algorithm training, and user modeling, emphasizing the crucial connections between these uses. The tutorial covers key simulation methodologies, with a particular focus on recent advancements leveraging large language models, discussing both the opportunities they present and the open challenges they entail. Crucially, we will also provide practical guidance, highlighting relevant toolkits, libraries, and datasets available to researchers and practitioners."
},
{
"venue": "CHIIR",
"title": "Search+Chat: Integrating Search and GenAI to Support Users with Learning-oriented Search Tasks",
"authors": [
"Yuyu Yang",
"Kelsey Urgo",
"Jaime Arguello",
"R. Capra"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3698204.3716446",
"source": "openalex",
"doi": "https://doi.org/10.1145/3698204.3716446",
"abstract": "Generative AI (GenAI) technologies such as ChatGPT are changing the ways people interact with information.To illustrate, popular search engines (e.g., Google) have started integrating responses from GenAI tools with the traditional search results.In this paper, we explore the integration of GenAI technology with traditional search in the context of a learning-oriented task.We report on a between-subjects study ( = 40) in which participants completed a complex, learning-oriented search task.Participants were assigned to one of two conditions.In the SearchOnly condition, participants used a traditional web search system to gather information.In the Search+Chat condition, participants used an experimental system that combined a traditional web search component and an interactive GenAI-based chat component (Chat AI).The study investigated seven research questions.RQ1-RQ3 focused on differences between groups: (RQ1) post-task perceptions, (RQ2) search behaviors, and (RQ3) learning outcomes.To measure learning, participants completed a multiple-choice test before the search task, immediately after, and one week later (to measure retention).RQ4-RQ7 delved deeper into participants' behaviors and experiences in the Search+Chat condition: (RQ4) motivations for (and gains from) engaging with the Chat AI; (RQ5) the phases during which participants engaged with the Chat AI; (RQ6) the types of queries issued to each component; and (RQ7) perceptions about the information returned by each component."
},
{
"venue": "CHIIR",
"title": "Design Principles for Exploratory Search Interfaces",
"authors": [
"Orland Hoeber"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3698204.3716443",
"source": "openalex",
"doi": "https://doi.org/10.1145/3698204.3716443",
"abstract": "Exploratory search has been proposed as a model of search behaviour that is well suited to complex search scenarios.However, the simple interfaces that are commonplace across many search contexts limit the ability for searchers to undertake exploratory searches.Little support is provided for the discovery, learning, and investigation necessary for exploratory browsing, or the query (re)formulation, result examination, and information extraction required for focused searching.While the design and study of search interfaces that accommodate and support searchers in undertaking exploratory searches has increased in recent years, much of this work has been ad hoc in nature.In this perspective paper, five search interface design principles are presented that are specifically tuned to support exploratory search.An extension of the classical heuristic evaluation method is provided to support the inspection of prototype search interfaces with respect to the design principles.Recent research in the field is categorized according to these design principles.Patterns and gaps in the literature are identified, highlighting opportunities for further research on exploratory search interfaces.These principles provide a framework to guide the design and inspection of future search interfaces to support exploratory search, as well as a mechanism for comparing and contrasting the interactive information retrieval literature as it relates to supporting exploratory search through novel interface design."
},
{
"venue": "CHIIR",
"title": "Assessing political bias and value misalignment in generative artificial intelligence",
"authors": [
"Fábio Motoki",
"Valdemar Pinho Neto",
"Victor Rangel"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.jebo.2025.106904",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.jebo.2025.106904",
"abstract": "Our analysis reveals a concerning misalignment of values between ChatGPT and the average American. We also show that ChatGPT displays political leanings when generating text and images, but the degree and direction of skew depend on the theme. Notably, ChatGPT repeatedly refused to generate content representing certain mainstream perspectives, citing concerns over misinformation and bias. As generative AI systems like ChatGPT become ubiquitous, such misalignment with societal norms poses risks of distorting public discourse. Without proper safeguards, these systems threaten to exacerbate societal divides and depart from principles that underpin free societies. • GPT-4’s responses align more with left-wing than average American political values. • AI text generation exhibits varying bias strength/direction across different themes. • Right-wing image generation refusals suggest potential First Amendment issues. • Meta-prompting induced GPT-4 to generate refused images without offensive content. • Calls for transparency, accountability from developers to mitigate societal risks."
},
{
"venue": "CHIIR",
"title": "Multi-stage Large Language Model Pipelines Can Outperform GPT-4o in Relevance Assessment",
"authors": [
"Julian A. Schnabel",
"Johanne R. Trippas",
"Falk Scholer",
"Danula Hettiachchi"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3701716.3715488",
"source": "openalex",
"doi": "https://doi.org/10.1145/3701716.3715488",
"abstract": "The effectiveness of search systems is evaluated using relevance labels that indicate the usefulness of documents for specific queries and users.While obtaining these relevance labels from real users is ideal, scaling such data collection is challenging.Consequently, third-party annotators are employed, but their inconsistent accuracy demands costly auditing, training, and monitoring.We propose an LLM-based modular classification pipeline that divides the relevance assessment task into multiple stages, each utilising different prompts and models of varying sizes and capabilities.Applied to TREC Deep Learning (TREC-DL), one of our approaches showed an 18.4% Krippendorff's accuracy increase over OpenAI's GPT-4o mini while maintaining a cost of about 0.2 USD per million input tokens, offering a more efficient and scalable solution for relevance assessment.This approach beats the baseline performance of GPT-4o (5 USD).With a pipeline approach, even the accuracy of the GPT-4o flagship model, measured in , could be improved by 9.7%."
},
{
"venue": "CHIIR",
"title": "DiSCo: LLM Knowledge Distillation for Efficient Sparse Retrieval in Conversational Search",
"authors": [
"Simon Lupart",
"Mohammad Aliannejadi",
"Evangelos Kanoulas"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729966",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729966",
"abstract": "Conversational Search (CS) involves retrieving relevant documents from a corpus while considering the conversational context, integrating retrieval with context modeling. Recent advancements in Large Language Models (LLMs) have significantly enhanced CS by enabling query rewriting based on conversational context. However, employing LLMs during inference poses efficiency challenges. Existing solutions mitigate this issue by distilling embeddings derived from human-rewritten queries, focusing primarily on learning the context modeling task. These methods, however, often separate the contrastive retrieval task from the distillation process, treating it as an independent loss term. To overcome these limitations, we introduce DiSCo (Distillation of Sparse Conversational retrieval), a novel approach that unifies retrieval and context modeling through a relaxed distillation objective. Instead of relying exclusively on representation learning, our method distills similarity scores between conversations and documents, providing more freedom in the representation space and better leveraging the contrastive nature of document relevance. Extensive experiments on Learned Sparse Retrieval (LSR) across five CS datasets demonstrate that DiSCo achieves substantial improvements in both in-domain and out-of-domain retrieval tasks, achieving up to a six-point gain in recall for out-of-domain datasets over state-of-the-art methods. Additionally, DiSCo employs a multi-teacher distillation strategy, using multiple LLMs as teachers, further enhancing performance and surpassing the individual teachers in in-domain settings. Furthermore, analysis of model sparsity reveals that DiSCo allows for more effective control over the sparsity of the trained models."
},
{
"venue": "CHIIR",
"title": "AI Mimicry and Human Dignity: Chatbot Use as a Violation of Self‐Respect",
"authors": [
"Jan‐Willem van der Rijt",
"Dimitri Coelho Mollo",
"Bram Vaassen"
],
"year": 2025,
"pdf_url": "https://onlinelibrary.wiley.com/doi/pdfdirect/10.1111/japp.70037",
"source": "openalex",
"doi": "https://doi.org/10.1111/japp.70037",
"abstract": "ABSTRACT This article investigates how human interactions with AI‐powered chatbots may offend human dignity. Current chatbots, driven by large language models, mimic human linguistic behaviour but lack the moral and rational capacities essential for genuine interpersonal respect. Human beings are prone to anthropomorphize chatbots – indeed, chatbots appear to be deliberately designed to elicit that response. As a result, human beings' behaviour towards chatbots often resembles behaviours typical of interaction between moral agents. Drawing on a second‐personal, relational account of dignity, we argue that interacting with chatbots in this way is incompatible with the dignity of users. We show that, since second‐personal respect is premised on reciprocal recognition of second‐personal moral authority, behaving towards chatbots in ways that convey second‐personal respect is bound to misfire in morally problematic ways, given the lack of reciprocity. Consequently, such chatbot interactions amount to subtle but significant violations of self‐respect – the respect we are duty‐bound to show for our own dignity. We illustrate this by discussing four actual chatbot use cases (information retrieval, customer service, advising, and companionship), and propound that the increasing societal pressure to engage in such interactions with chatbots poses a hitherto underappreciated threat to human dignity."
},
{
"venue": "CHIIR",
"title": "From Query to Conscience: The Importance of Information Retrieval in Empowering Socially Responsible Consumerism",
"authors": [
"Frans van der Sluis",
"Leif Azzopardi",
"Florian Meier"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730347",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730347",
"abstract": "Millions of consumers search for products online each day, aiming to find items that meet their needs at an acceptable price. While price and quality are major factors in purchasing decisions, ethical considerations increasingly influence consumer behavior -giving rise to the socially responsible consumer. Insights from a recent survey of over 600 consumers reveal that many barriers to ethical shopping stem from information-seeking challenges, often leading to decisions made under uncertainty. These challenges contribute to the intention-behaviour gap, where consumers' desire to make ethical choices is undermined by limited or inaccessible information and inefficacy of search systems in supporting responsible decision-making. In this perspectives paper, we argue that the field of Information Retrieval (IR) has a critical role to play by empowering consumers to make more informed and more responsible choices. We present three interrelated perspectives: (1) reframing ethical consumption as an information extraction problem aimed at reducing information asymmetries; (2) redefining product search as a complex task requiring interfaces that lower the cost and burden of responsible search; and (3) reimagining search as a process of knowledge calibration that helps consumers bridge gaps in awareness when making purchasing decisions. Taken together, these perspectives outline a path from query to conscience - one where IR systems help transform everyday product searches into opportunities for more ethical and informed choices. We advocate for the development of new and novel IR systems and interfaces that address the intricacies of socially responsible consumerism, and call on the IR community to build technologies that make ethical decisions more informed, convenient, and aligned with economic realities."
},
{
"venue": "CHIIR",
"title": "Academic Advising Chatbot Powered with AI Agent",
"authors": [
"Michael Tamascelli",
"Olivia Bunch",
"Blake Fowler",
"Maryam Taeb",
"Achraf Cohen"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696673.3723065",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696673.3723065",
"abstract": "Academic advising plays a crucial role in fostering student success. However, challenges such as limited advisor availability can hinder effective support. Generative AI, particularly AI-powered chatbots, offers the potential to enhance student advising in higher education by providing personalized guidance. These technologies help college students find the information and resources needed to create degree plans aligned with their academic goals. This research introduces ARGObot, an intelligent advising system that facilitates student navigation of university policies through automated interpretation of the student handbook as its primary knowledge base. ARGObot enhances accessibility to critical academic policies and procedures, supporting incoming students' success through personalized guidance. Our system integrates a multifunctional agent enhanced by a Large Language Model (LLM). The architecture employs multiple external tools to enhance its capabilities: a Retrieval-Augmented Generation (RAG) system accesses verified university sources; email integration facilitates Human-in-the-Loop (HITL) interaction; and a web search function expands the system's knowledge base beyond predefined constraints. This approach enables the system to provide contextually relevant and verified responses to various student queries. This architecture evolved from our initial implementation based on Gemini 1 Pro, which revealed significant limitations due to its lack of agent-based functionality, resulting in hallucination issues and irrelevant responses. Subsequent evaluation demonstrated that our enhanced version, integrating GPT-4 with the text-embedding-ada-002 model, achieved superior performance across all metrics. This paper also presents a comparative analysis of both implementations, highlighting the architectural improvements and their impact on system performance."
},
{
"venue": "CHIIR",
"title": "From Links to Dialogue; Hypertext Challenges and Opportunities in Conversational Navigation",
"authors": [
"Behnam Rahdari",
"Peter Brusilovsky"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3720533.3750064",
"source": "openalex",
"doi": "https://doi.org/10.1145/3720533.3750064",
"abstract": "Large Language Model (LLM) dialogue is rapidly displacing the blue-link lists that once defined web navigation.While LLM with its fluent answers delights users, this shift affects readers' agency concealing the associative links and orienting cues that the hypertext community has spent eight decades refining.This paper asks: What must be reclaimed, and what new affordances are possible, when navigation is mediated by a conversational model?We revisit seminal systems from Bush's Memex to Intermedia and Storyspace to surface five core principles (associative linking, agency, information scent, non-linearity, maps) and trace how each is strained or obscured in single-pane chat.By framing the problem through humandata interaction, we articulate three design obligations-legibility, agency, and negotiability-and demonstrate how emerging techniques such as evidence cards, trail-map overlays, diversity sliders, and on-device tiny LLMs can move us towards reclaiming user agency in navigating the web.We then outline a research agenda that ranges from authoring grammars for LLM-mediated hypertexts to ethical navigation standards that curb bias and filter bubbles.Near-term tweaks are actionable today; longer-term questions chart a collaborative path for hypertext, HCI, and AI researchers.Combining classic hypertext insights with modern LLM capabilities, we aim to outline a road map for conversational interfaces that preserve critical reading and empower users to see (and steer) the trails behind every answer."
},
{
"venue": "CHIIR",
"title": "Is Relevance Propagated from Retriever to Generator in RAG?",
"authors": [
"Feng Tian",
"Debasis Ganguly",
"Craig Macdonald"
],
"year": 2025,
"pdf_url": "https://eprints.gla.ac.uk/343843/3/343843.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/978-3-031-88708-6_3",
"abstract": ""
},
{
"venue": "CHIIR",
"title": "How Users Interact with Generative Information Retrieval Systems: A Study of User Behavior and Search Experience",
"authors": [
"Yuming Liang",
"Zhijing Wu",
"Fan Zhang",
"Dandan Song",
"Heyan Huang"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729998",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729998",
"abstract": "The development of LLM has facilitated the emergence of generative information retrieval (IR) systems, such as ''Bing Chat''. Generative IR systems return generated text with citations rather than a list of ranked search results. User studies on IR systems are essential for understanding users' interaction patterns, evaluating and optimizing systems, and improving search experience, particularly in the context of generative IR systems with novel conversational interfaces and responses. However, systematic investigations into user behavior and search experience on generative IR systems are notably lacking. To address this gap, we conducted a user study using Bing Chat to explore user behavior and feedback on generative IR systems. The participants were required to accomplish three types of tasks using Bing Chat. During the search process, we collected their various behavior (e.g., click, query reformulation) and explicit feedback (e.g., satisfaction, credibility, and success). Additionally, the same study was conducted on traditional IR systems Bing for comparison. Analyses of these data show that Bing Chat can reduce the user's search effort and lead to a better search experience without any decrease in credibility compared with Bing. We believe that this work provides valuable insight into the design and evaluation of generative information retrieval systems."
},
{
"venue": "CHIIR",
"title": "Applying Large Language Models to Interactive Information Retrieval: A Practical Exploration",
"authors": [
"Johanne R. Trippas",
"Oleg Zendel",
"Adam Roegiest"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3698204.3716478",
"source": "openalex",
"doi": "https://doi.org/10.1145/3698204.3716478",
"abstract": "This half-day interactive tutorial provides researchers with practical skills to use large language models (LLMs) for interactive information retrieval research.Through hands-on exercises and real-world research examples, participants will learn to set up LLMs locally, integrate them via APIs, and evaluate their outputs.The tutorial will explain which models suit specific research needs, offering participants a robust toolkit to enhance their work.Attendees will also gain insights into the latest developments in the field, ensuring they stay at the forefront of innovation.Ideal for researchers eager to explore new methodologies in information retrieval, this tutorial offers foundational knowledge and cutting-edge strategies to use LLMs in interactive information retrieval research."
},
{
"venue": "CHIIR",
"title": "SOLID: Self-seeding and Multi-intent Self-instructing LLMs for Generating Intent-aware Information-Seeking Dialogs",
"authors": [
"Arian Askari",
"Roxana Petcu",
"Chuan Meng",
"Mohammad Aliannejadi",
"Amin Abolghasemi",
"Evangelos Kanoulas",
"Suzan Verberne"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-naacl.357.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-naacl.357",
"abstract": "Intent prediction in information-seeking dialogs is challenging and requires a substantial amount of data with human-labeled intents for effective model training.While Large Language Models (LLMs) have demonstrated effectiveness in generating synthetic data, existing methods typically rely on human feedback and are tailored to structured, task-oriented intents.In this paper, we leverage LLMs for zero-shot generation of large-scale, opendomain, intent-aware information-seeking dialogs to serve as training data for intent prediction models.We introduce SOLID, a method that generates dialogs turn by turn using novel self-seeding and multi-intent self-instructing strategies.Additionally, we propose SOLID-RL, a finetuned version that generates an entire dialog in one step using data created with SOLID.SOLID and SOLID-RL are each used to generate over 300k intent-aware dialogs, significantly surpassing the size of existing datasets.Experiments show that intent prediction models trained on sampled dialogs generated by SOLID and SOLID-RL outperform those trained solely on human-generated dialogs.Our findings demonstrate the potential of LLMs to expand training datasets, as they provide valuable resources for conversational agents across multiple tasks.Our self-seeding and self-instructing approaches are adaptable to various conversational data types and languages with minimal modifications.* * Equal contribution (shared co-authorship)Can you provide details on integrating Slack with other messengers and cloud apps?"
},
{
"venue": "CHIIR",
"title": "Unraveling the Impact of Visual Complexity on Search as Learning",
"authors": [
"Wolfgang Gritz",
"Anett Hoppe",
"Ralph Ewerth"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2501.05289",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2501.05289",
"abstract": "Information search has become essential for learning and knowledge acquisition, offering broad access to information and learning resources. The visual complexity of web pages is known to influence search behavior, with previous work suggesting that searchers make evaluative judgments within the first second on a page. However, there is a significant gap in our understanding of how visual complexity impacts searches specifically conducted with a learning intent. This gap is particularly relevant for the development of optimized information retrieval (IR) systems that effectively support educational objectives. To address this research need, we model visual complexity and aesthetics via a diverse set of features, investigating their relationship with search behavior during learning-oriented web sessions. Our study utilizes a publicly available dataset from a lab study where participants learned about thunderstorm formation. Our findings reveal that while content relevance is the most significant predictor for knowledge gain, sessions with less visually complex pages are associated with higher learning success. This observation applies to features associated with the layout of web pages rather than to simpler features (e.g., number of images). The reported results shed light on the impact of visual complexity on learning-oriented searches, informing the design of more effective IR systems for educational contexts. To foster reproducibility, we release our source code (https://github.com/TIBHannover/sal_visual_complexity)."
},
{
"venue": "CHIIR",
"title": "Tip of the Tongue Query Elicitation for Simulated Evaluation",
"authors": [
"Yifan He",
"To Eun Kim",
"Fernando Díaz",
"Jaime Arguello",
"Bhaskar Mitra"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730335",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730335",
"abstract": "Tip-of-the-tongue (TOT) search occurs when a user struggles to recall a specific identifier, such as a document title. While common, existing search systems often fail to effectively support TOT scenarios. Research on TOT retrieval is further constrained by the challenge of collecting queries, as current approaches rely heavily on community question-answering (CQA) websites, leading to labor-intensive evaluation and domain bias. To overcome these limitations, we introduce two methods for eliciting TOT queries-leveraging large language models (LLMs) and human participants-to facilitate simulated evaluations of TOT retrieval systems. Our LLM-based TOT user simulator generates synthetic TOT queries at scale, achieving high correlations with how CQA-based TOT queries rank TOT retrieval systems when tested in the Movie domain. Additionally, these synthetic queries exhibit high linguistic similarity to CQA-derived queries. For human-elicited queries, we developed an interface that uses visual stimuli to place participants in a TOT state, enabling the collection of natural queries. In the Movie domain, system rank correlation and linguistic similarity analyses confirm that human-elicited queries are both effective and closely resemble CQA-based queries. These approaches reduce reliance on CQA-based data collection while expanding coverage to underrepresented domains, such as Landmark and Person. LLM-elicited queries for the Movie, Landmark, and Person domains have been released as test queries in the TREC 2024 TOT track, with human-elicited queries scheduled for inclusion in the TREC 2025 TOT track. Additionally, we provide source code for synthetic query generation and the human query collection interface, along with curated visual stimuli used for eliciting TOT queries."
},
{
"venue": "CHIIR",
"title": "Guarding Our Vital Systems: A Metric for Critical Infrastructure Cyber Resilience",
"authors": [
"Muharman Lubis",
"Muhammad Fakhrul Safitra",
"Hanif Fakhrurroja",
"Alif Noorachmad Muttaqin"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/1424-8220/25/15/4545/pdf?version=1753191923",
"source": "openalex",
"doi": "https://doi.org/10.3390/s25154545",
"abstract": "The increased occurrence and severity of cyber-attacks on critical infrastructure have underscored the need to embrace systematic and prospective approaches to resilience. The current research takes as its hypothesis that the InfraGuard Cybersecurity Framework-a capability model that measures the maturity of cyber resilience through three functional pillars, Cyber as a Shield, Cyber as a Space, and Cyber as a Sword-is an implementable and understandable means to proceed with. The model treats the significant aspects of situational awareness, active defense, risk management, and recovery from incidents and is measured using globally standardized maturity models like ISO/IEC 15504, NIST CSF, and COBIT. The contributions include multidimensional measurements of resilience, a scored scale of capability (0-5), and domain-based classification enabling organizations to assess and enhance their cybersecurity situation in a formalized manner. The framework's applicability is illustrated in three exploratory settings of power grids, healthcare systems, and airports, each constituting various levels of maturity in resilience. This study provides down-to-earth recommendations to policymakers through the translation of the attributes of resilience into concrete assessment indicators, promoting policymaking, investment planning, and global cyber defense collaboration."
},
{
"venue": "CHIIR",
"title": "Exploring the Zero-Shot Known-Item Retrieval Capabilities of LLMs for Casual Leisure Information Needs",
"authors": [
"Toine Bogers",
"Maria Gäde",
"Mark Hall",
"Marijn Koolen",
"Vivien Petras",
"Mette Skov"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3698204.3716466",
"source": "openalex",
"doi": "https://doi.org/10.1145/3698204.3716466",
"abstract": "The rapidly increasing popularity of LLM-powered chatbots has led to them being used for a increasing number of different tasks by the general public.One of these tasks is searching for information instead of using a search engine.Previous work has shown that complex search tasks can be problematic for traditional search engines to solve, but little is known about the capability of LLMs on the same task.We compared four LLMs on their capability to answer a specific type of complex search task: known-item requests from casual leisure domains.We constructed a test collection by gathering known-item requests for books, games and movies from online forums along with verified answers by the original requester.We prompted four LLMs multiple times with the same prompt and analyzed the results with respect to accuracy and the degree to which answers were fabricated by the LLM.Our results show that LLMs are not particularly effective in fulfilling these complex casual leisure needs, but there are are big differences between LLMs and across domains. CCS Concepts"
},
{
"venue": "CHIIR",
"title": "The Viability of Crowdsourcing for RAG Evaluation",
"authors": [
"Lukas Gienapp",
"Thomas Hagen",
"Maik Fröbe",
"Matthias Hagen",
"Benno Stein",
"Martin Potthast",
"Harrisen Scells"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730093",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730093",
"abstract": "How good are humans at writing and judging responses in retrieval-augmented generation (RAG) scenarios? To answer this question, we investigate the efficacy of crowdsourcing for RAG through two complementary studies: response writing and response utility judgment. Our new Webis Crowd RAG Corpus 2025 (Webis-CrowdRAG-25) consists of 903 human-written and 903 LLM-generated responses for the 301 topics of the TREC 2024 RAG~track, with each response composed according to one of the three discourse styles 'bullet list', 'essay', or 'news'. For a selection of 65 topics, the corpus further contains 47,320 pairwise human judgments and 10,556 pairwise LLM judgments across seven utility dimensions (e.g., coverage and coherence). Our analyses give insights into human writing behavior for RAG and the viability of crowdsourcing for RAG evaluation. We find that human pairwise judgments provide reliable and cost-effective results. This is much less the case for LLM-based pairwise and human/LLM-based pointwise judgments, nor for automated comparisons with human-written reference responses. All our data and tools are freely available."
},
{
"venue": "CHIIR",
"title": "Bridging the Gap: From Ad-hoc to Proactive Search in Conversations",
"authors": [
"Chuan Meng",
"Francesco Tonolini",
"Fengran Mo",
"Νικόλαος Αλέτρας",
"Emine Yilmaz",
"Gabriella Kazai"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729915",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729915",
"abstract": "Proactive search in conversations (PSC) aims to reduce user effort in formulating explicit queries by proactively retrieving useful relevant information given conversational context. Previous work in PSC either directly uses this context as input to off-the-shelf ad-hoc retrievers or further fine-tunes them on PSC data. However, ad-hoc retrievers are pre-trained on short and concise queries, while the PSC input is longer and noisier. This input mismatch between ad-hoc search and PSC limits retrieval quality. While fine-tuning on PSC data helps, its benefits remain constrained by this input gap. In this work, we propose Conv2Query, a novel conversation-to-query framework that adapts ad-hoc retrievers to PSC by bridging the input gap between ad-hoc search and PSC. Conv2Query maps conversational context into ad-hoc queries, which can either be used as input for off-the-shelf ad-hoc retrievers or for further fine-tuning on PSC data. Extensive experiments on two PSC datasets show that Conv2Query significantly improves ad-hoc retrievers' performance, both when used directly and after fine-tuning on PSC."
},
{
"venue": "CHIIR",
"title": "From To-Do to Ta-Da: Transforming Task-Focused IR with Generative AI",
"authors": [
"Chirag Shah",
"Ryen W. White"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730352",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730352",
"abstract": "For decades, scholars have emphasized that tasks should be the central focus in Information Retrieval (IR).This point of view holds even more significance with the advent of Generative Artificial Intelligence (GenAI) models, which can, among other capabilities, understand natural language, engage in dialog with users, generate bespoke user interfaces, and power agents to help complete tasks.GenAI presents an unprecedented opportunity to finally realize the potential of tasks in IR, enhance task-focused retrieval and interaction, and create \"magical\" task completion moments for users.In this paper, we explore the rationale and methodology behind this argument.Traditional IR systems support mostly simple tasks.The emergence of GenAI creates an opportunity for IR systems to help users achieve complex tasks and for the IR community to rekindle its interest and demonstrate leadership in this sizable and significant problem space.We underscore the pivotal role of tasks in IR and introduce new evidence supporting the notion that task-centric approaches, abstracted from specific modalities, represent the future of IR.Building on this foundation, we envision the development, utilization, and evaluation of next-generation IR systems.We propose a promising future where IR agents prioritize users, their tasks, and their situations.However, despite their potential to address task-focused and modality-independent IR, agents alone are insufficient.We propose a robust ecosystem around these agents that transcends traditional queries, questions, prompts, and modalities to address users' fundamental needs, tasks, and goals."
},
{
"venue": "CHIIR",
"title": "Emerging Trends in Information Science: Latest advancement and future directions in libraries",
"authors": [
"Balqis Basyirah Burhan",
"Saiful Farik Mat Yatin",
"Iffah Farhana Mustafa",
"Nur Diyana Sofiah Erwanis",
"Nuratiqkah Harith",
"Roslaizam Daud"
],
"year": 2025,
"pdf_url": "https://ebpj.e-iph.co.uk/index.php/EBProceedings/article/download/6830/4269",
"source": "openalex",
"doi": "https://doi.org/10.21834/e-bpj.v10isi27.6830",
"abstract": "Libraries are transforming in the digital era, evolving from knowledge repositories into hubs of technology, education, and community. Among the latest advancements, VR/AR enables immersive learning, AI enhances personalized services, and blockchain secures digital assets. While these innovations reshape libraries, challenges like budget constraints and the digital divide persist, especially in rural areas. Overcoming these requires strategic partnerships, funding, and inclusivity-focused initiatives. This article highlights libraries’ evolution in leveraging technology to ensure equitable access. Future directions include adopting digital ecosystems, utilizing AI-driven analytics, and expanding VR/AR for collaborative learning, cementing libraries as pivotal in a technology-driven world."
},
{
"venue": "CHIIR",
"title": "A Flexible User Study Platform for Generative Information Retrieval",
"authors": [
"Yuming Liang",
"Zhijing Wu",
"Yuchen He",
"Fengming Liang",
"Kexin Liu",
"Jiaxin Mao"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730140",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730140",
"abstract": "User behavior and experience are important for improving information retrieval (IR) systems. While much research has focused on traditional IR systems, few studies have systematically examined user behavior and search experience with emerging generative IR systems. A key reason for this gap is the lack of publicly available toolkits to record user behavior and feedback in generative IR systems. We developed a comprehensive platform to collect user behavior and feedback on the generative IR system. This platform consists of: 1) a generative IR system that supports both API-based and customized retrieval-augmented generation (RAG) methods, 2) a user interface that logs various user behavior, including prompts, clicks, mouse movements, and scrolling, and 3) an annotation website that allows users to provide feedback. We believe the proposed platform has the potential to streamline data collection for user studies on generative IR systems, paving the way for future research on how users engage with and interact with these systems."
},
{
"venue": "CHIIR",
"title": "Charting the Landscape of Artificial Intelligence Ethics: A Bibliometric Analysis",
"authors": [
"Jiaxuan Qiu",
"Le Cheng",
"Jin Huang"
],
"year": 2025,
"pdf_url": "https://www.degruyterbrill.com/document/doi/10.1515/ijdlg-2025-0007/pdf",
"source": "openalex",
"doi": "https://doi.org/10.1515/ijdlg-2025-0007",
"abstract": "Abstract Using bibliometric methods, this study systematically analyzes 6,084 AI ethics-related articles from the Web of Science Core Collection (2015–2025), capturing both recent advances and near-future directions in the field. It begins by examining publication trends, disciplinary categories, leading journals, and major contributing institutions/countries. Subsequently, co-citation (journals, authors, references) and keyword clustering methods reveal the foundational knowledge structure and highlight emerging research hotspots. The findings indicate increasing interdisciplinary convergence and international collaboration in AI ethics, with core themes focusing on algorithmic fairness, privacy and data security, ethical governance in autonomous vehicles, medical AI applications, educational technology, and challenges posed by generative AI (e.g., large language models). Burst keyword detection further shows an evolutionary shift from theoretical debates toward practical implementation strategies and regulatory framework development. Although numerous global initiatives have been introduced to guide AI ethics, broad consensus remains elusive, underscoring the need for enhanced cross-disciplinary and international cooperation. This research provides valuable insights for scholars, policymakers, and industry practitioners, laying a foundation for sustainable and responsible AI development."
},
{
"venue": "CHIIR",
"title": "Clarifying Ambiguities: on the Role of Ambiguity Types in Prompting Methods for Clarification Generation",
"authors": [
"Anfu Tang",
"Laure Soulier",
"Vincent Guigue"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729922",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729922",
"abstract": "In information retrieval (IR), providing appropriate clarifications to better understand users' information needs is crucial for building a proactive search-oriented dialogue system. Due to the strong in-context learning ability of large language models (LLMs), recent studies investigate prompting methods to generate clarifications using few-shot or Chain of Thought (CoT) prompts. However, vanilla CoT prompting does not distinguish the characteristics of different information needs, making it difficult to understand how LLMs resolve ambiguities in user queries. In this work, we focus on the concept of ambiguity for clarification, seeking to model and integrate ambiguities in the clarification process. Following the reasoning and acting paradigm, we propose a new prompting scheme Ambiguity Type-Chain of Thought (AT-CoT), which enhances the reasoning abilities of LLMs by limiting CoT to first predict ambiguity types that can be interpreted as actions, then generate clarifications correspondingly. Experiments are conducted on various datasets containing human-annotated clarifying questions to compare AT-CoT with multiple baselines. We also perform user simulation to implicitly measure the quality of generated clarifications under various IR scenarios. Our codes are available at: https://github.com/anfutang/ClarifyingAmbiguities/."
},
{
"venue": "CHIIR",
"title": "LLM4Eval: Large Language Model for Evaluation in IR",
"authors": [
"Clemencia Siro",
"Hossein A. Rahmani",
"Mohammad Aliannejadi",
"Nick Craswell",
"Charles L. A. Clarke",
"Guglielmo Faggioli",
"Bhaskar Mitra",
"Paul Thomas",
"Emine Yilmaz"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730367",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730367",
"abstract": "Large language models (LLMs) have demonstrated increasing task-solving abilities not present in smaller models. Utilizing the capabilities and responsibilities of LLMs for automated evaluation (LLM4Eval) has recently attracted considerable attention in multiple research communities. Building on the success of previous workshops, which established foundations in automated judgments and RAG evaluation, this third iteration aims to address emerging challenges as IR systems become increasingly personalized and interactive. The main goal of the third LLM4Eval workshop is to bring together researchers from industry and academia to explore three critical areas: the evaluation of personalized IR systems while maintaining fairness, the boundaries between automated and human assessment in subjective scenarios, and evaluation methodologies for systems that combine multiple IR paradigms (search, recommendations, and dialogue). By examining these challenges, we seek to understand how evaluation approaches can evolve to match the sophistication of modern IR applications. The format of the workshop is interactive, including roundtable discussion sessions, fostering dialogue about the future of IR evaluation while avoiding one-sided discussions. This is the third iteration of the workshop series, following successful events at SIGIR 2024 and WSDM 2025, with the first iteration attracting over 50 participants."
},
{
"venue": "CHIIR",
"title": "Federated Binary Matrix Factorization Using Proximal Optimization",
"authors": [
"Sebastian Dalleiger",
"Jilles Vreeken",
"Michael Kamp"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/33773/35928",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i15.33773",
"abstract": "Identifying informative components in binary data is an essential task in many application areas, including life sciences, social sciences, and recommendation systems. Boolean matrix factorization (BMF) is a family of methods that performs this task by factorizing the data into dense factor matrices. In real-world settings, the data is often distributed across stakeholders and required to stay private, prohibiting the straightforward application of BMF. To adapt BMF to this context, we approach the problem from a federated-learning perspective, building on a state-of-the-art continuous binary matrix factorization relaxation to BMF that enables efficient gradient-based optimization. Our approach only needs to share the relaxed component matrices, which are aggregated centrally using a proximal operator that regularizes for binary outcomes. We show the convergence of our federated proximal gradient descent algorithm and provide differential privacy guarantees. Our extensive empirical evaluation shows that our algorithm outperforms, in quality and efficacy, federation schemes of state-of-the-art BMF methods on a diverse set of real-world and synthetic data."
},
{
"venue": "CHIIR",
"title": "LLM-Assisted Relevance Assessments: When Should We Ask LLMs for Help?",
"authors": [
"Rikiya Takehi",
"Ellen M. Voorhees",
"Tetsuya Sakai",
"Ian Soboroff"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729916",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729916",
"abstract": "Test collections are information retrieval tools that allow researchers to quickly and easily evaluate ranking algorithms. While test collections have become an integral part of IR research, the process of data creation involves significant manual annotation effort, which often makes it very expensive and time-consuming. Consequently, test collections could become too small when the budget is limited, which may lead to unstable evaluations. As a cheaper alternative, recent studies have proposed the use of large language models (LLMs) to completely replace human assessors. However, while LLMs seem to somewhat correlate with human judgments, their predictions are not perfect and often show bias. Thus, a complete replacement with LLMs is argued to be too risky and not fully reliable."
},
{
"venue": "CHIIR",
"title": "Small Data, Big Impact: Navigating Resource Limitations in Point-of-Interest Recommendation for Individuals with Autism",
"authors": [
"Ludovico Boratto",
"Federica Cena",
"Mirko Marras",
"Noemi Mauro",
"Giacomo Medda"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730269",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730269",
"abstract": "Autism Spectrum Disorder (ASD) affects sensory perception, making spatial exploration difficult. Recommender systems can assist ASD users by suggesting Points of Interest (POIs) aligned with their sensory preferences. However, demographic constraints, difficulties in engaging ASD users, and the complexity of obtaining sensory data position POI recommendation for ASD people as a low-resource problem. In this paper, we identify key challenges in developing such systems and present our ongoing efforts. Using a local ASD center as a use case, we are developing a structured user involvement protocol. From the limited data, we are deriving knowledge graphs (KGs) to model preferences and sensory aspects. We are then exploring KG-based techniques to generate paths from users to POIs to suggest. With psychologists, we are refining the paths structure to match varying complexity levels and translate them into natural language accessible for people with ASD."
},
{
"venue": "CHIIR",
"title": "A Versatile Dataset of Mouse and Eye Movements on Search Engine Results Pages",
"authors": [
"Kayhan Latifzadeh",
"Jacek Gwizdka",
"Luis A. Leiva"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730325",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730325",
"abstract": "We contribute a comprehensive dataset to study user attention and purchasing behavior on Search Engine Result Pages (SERPs). Previous work has relied on mouse movements as a low-cost large-scale behavioral proxy but also has relied on self-reported ground-truth labels, collected at post-task, which can be inaccurate and prone to biases. To address this limitation, we use an eye tracker to construct an objective ground-truth of continuous visual attention. Our dataset comprises 2,776 transactional queries on Google SERPs, collected from 47 participants, and includes: (1)~HTML source files, with CSS and images; (2)~rendered SERP screenshots; (3)~eye movement data; (4)~mouse movement data; (5)~bounding boxes of direct display and organic advertisements; and (6)~scripts for further preprocessing the data. In this paper we provide an overview of the dataset and baseline experiments (classification tasks) that can inspire researchers about the different possibilities for future work."
},
{
"venue": "CHIIR",
"title": "Are Generative AI Agents Effective Personalized Financial Advisors?",
"authors": [
"Takehiro Takayanagi",
"Kiyoshi Izumi",
"Javier Sanz-Cruzado",
"Richard McCreadie",
"Iadh Ounis"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729897",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729897",
"abstract": "Large language model-based agents are becoming increasingly popular as a low-cost mechanism to provide personalized, conversational advice, and have demonstrated impressive capabilities in relatively simple scenarios, such as movie recommendations. But how do these agents perform in complex high-stakes domains, where domain expertise is essential and mistakes carry substantial risk? This paper investigates the effectiveness of LLM-advisors in the finance domain, focusing on three distinct challenges: (1) eliciting user preferences when users themselves may be unsure of their needs (2) providing personalized guidance for diverse investment preferences, and (3) leveraging advisor personality to build relationships and foster trust. Via a lab-based user study with 64 participants, we show that LLM-advisors often match human advisor performance when eliciting preferences, although they can struggle to resolve conflicting user needs. When providing personalized advice, the LLM was able to positively influence user behavior, but demonstrated clear failure modes. Our results show that accurate preference elicitation is key, otherwise, the LLM-advisor has little impact, or can even direct the investor toward unsuitable assets. More worryingly, users appear insensitive to the quality of advice being given, or worse these can have an inverse relationship. Indeed, users reported a preference for and increased satisfaction as well as emotional trust with LLMs adopting an extroverted persona, even though those agents provided worse advice."
},
{
"venue": "CHIIR",
"title": "PRESTO: A Recommender of Musical Collaborations Based on Heterogeneous Graph Neural Networks",
"authors": [
"Fernando Terroso-Sáenz",
"J. Soto",
"Andrés Muñoz",
"Philippe Roose"
],
"year": 2025,
"pdf_url": "https://www.ijimai.org/index.php/ijimai/article/download/863/922",
"source": "openalex",
"doi": "https://doi.org/10.9781/ijimai.2025.03.004",
"abstract": "The music industry is now more complex and competitive than ever before. In recent years, the search for collaborations with other artists has become a common strategy for musicians to maintain their presence in the sector. Besides, existing music streaming services such as Spotify have exposed large data feeds that can be used to develop innovative services within the realm of music. In this context, the present work introduces PRESTO, a novel recommendation system to suggest musicians for new collaborations with other artists by means of an ensemble of Graph Neural Networks. The system is fed with a heterogeneous graph representing the time evolution and the stationary aspects of a musician’s career. Finally, the proposal has been evaluated with a dataset comprising more than 200,000 artists, with an average F1 score above 0.75."
},
{
"venue": "CHIIR",
"title": "A study of search result aggregation approaches for the digital humanities",
"authors": [
"Milad Momeni",
"Orland Hoeber"
],
"year": 2025,
"pdf_url": "https://onlinelibrary.wiley.com/doi/pdfdirect/10.1002/asi.70006",
"source": "openalex",
"doi": "https://doi.org/10.1002/asi.70006",
"abstract": "Abstract Searching across diverse information platforms, such as digital humanities archives, academic digital libraries, and encyclopedias, poses challenges in managing the queries issued to each platform and synthesizing the resources discovered. While search result aggregation interfaces address this problem, how best to present the search results from different platforms in the search engine results page remains an open question. In this research, we implemented three common approaches and developed a new technique for aggregating search results across three platforms: Europeana, our University's academic library, and Wikipedia. The three common approaches (1) use tabs to switch between the platforms, (2) interleave results from each platform producing a single list, and (3) use a bento box approach to group results from each platform. The new technique organizes the search results into thematic clusters irrespective of their source platform. We designed a controlled laboratory study using a within‐subjects design and exploratory search tasks conducted in the context of digital humanities searching. We collected data from 32 student participants, focusing on utility, perceived value, and diversity of saved resources. This study provides evidence that thematic clustering can be a beneficial aggregation approach, opening opportunities for studying different ways of representing and visualizing aggregated search results."
},
{
"venue": "CHIIR",
"title": "Large Language Model Relevance Assessors Agree With One Another More Than With Human Assessors",
"authors": [
"Maik Fröbe",
"Andrew Parry",
"Ferdinand Schlatt",
"Sean MacAvaney",
"Benno Stein",
"Martin Potthast",
"Matthias Hagen"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730218",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730218",
"abstract": "Relevance judgments can differ between assessors, but previous work has shown that such disagreements have little impact on the effectiveness rankings of retrieval systems. This applies to disagreements between humans as well as between human and large language model (LLM) assessors. However, the agreement between different LLM~assessors has not yet been systematically investigated. To close this gap, we compare eight LLM~assessors on the TREC DL tracks and the retrieval task of the RAG track with each other and with human assessors. We find that the agreement between LLM~assessors is higher than between LLMs and humans and, importantly, that LLM~assessors favor retrieval systems that use LLMs in their ranking decisions: our analyses with 30-50 retrieval systems show that the system rankings obtained by LLM~assessors overestimate LLM-based re-rankers by 9~to 17~positions on average."
},
{
"venue": "CHIIR",
"title": "Do Images Clarify? A Study on the Effect of Images on Clarifying Questions in Conversational Search",
"authors": [
"Clemencia Siro",
"Zahra Abbasiantaeb",
"Yifei Yuan",
"Mohammad Aliannejadi",
"Maarten de Rijke"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3698204.3716464",
"source": "openalex",
"doi": "https://doi.org/10.1145/3698204.3716464",
"abstract": "Conversational search (CS) systems increasingly employ clarifying questions to refine user queries and improve the search experience.Previous studies have demonstrated the usefulness of text-based clarifying questions in enhancing both retrieval performance and user experience.While images have been shown to improve retrieval performance in various contexts, their impact on user performance, when incorporated into clarifying questions, remains largely unexplored.We conduct a user study with 73 participants to investigate the role of images in CS, specifically examining their effects on two search-related tasks: (i) answering clarifying questions, and (ii) query reformulation.We compare the effect of multimodal and text-only clarifying questions in both tasks within a CS context from various perspectives.Our findings reveal that while participants showed a strong preference for multimodal questions when answering clarifying questions, preferences were more balanced in the query reformulation task.The impact of images varied with both task type and user expertise: in answering clarifying questions, images helped maintain engagement across different expertise levels, while in query reformulation, they led to more precise queries and improved retrieval performance.Interestingly, for clarifying question answers, text-only setups demonstrated better user performance as they provided more comprehensive textual information in the absence of images.These results provide valuable insights for designing effective multimodal CS systems, highlighting that the benefits of visual augmentation are task-dependent and should be strategically implemented based on the specific search context and user characteristics."
},
{
"venue": "CHIIR",
"title": "Conversational Gold: Evaluating Personalized Conversational Search System Using Gold Nuggets",
"authors": [
"Zahra Abbasiantaeb",
"Simon Lupart",
"Leif Azzopardi",
"Jeff Dalton",
"Mohammad Aliannejadi"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730316",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730316",
"abstract": "The rise of personalized conversational search systems has been driven by advancements in Large Language Models (LLMs), enabling these systems to retrieve and generate answers for complex information needs. However, the automatic evaluation of responses generated by Retrieval Augmented Generation (RAG) systems remains an understudied challenge. In this paper, we introduce a new resource for assessing the retrieval effectiveness and relevance of responses generated by RAG systems, using a nugget-based evaluation framework. Built upon the foundation of TREC iKAT 2023, our dataset extends to the TREC iKAT 2024 collection, which includes 17 conversations and 20,575 relevance passage assessments, together with 2,279 extracted gold nuggets and 62 manually written gold answers from NIST assessors. While maintaining the core structure of its predecessor, this new collection enables a deeper exploration of generation tasks in conversational settings. Key improvements in iKAT 2024 include: (1) ''gold nuggets'' - concise, essential pieces of information extracted from relevant passages of the collection - which serve as a foundation for automatic response evaluation; (2) manually written answers to provide a gold standard for response evaluation; (3) expanded user personas, providing richer contextual grounding; and (4) a transition from Personal Text Knowledge Base (PTKB) ranking to PTKB classification and selection. Built on this resource, we provide a framework for long-form answer generation evaluation, involving nugget extraction and nugget matching, linked to retrieval. This establishes a solid resource for advancing research in personalized conversational search and long-form answer generation. Our resources are publicly available at https://github.com/irlabamsterdam/CONE-RAG."
},
{
"venue": "CHIIR",
"title": "Measuring Text-Image Retrieval Fairness with Synthetic Data",
"authors": [
"Lluís Gómez"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730030",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730030",
"abstract": "In this paper, we study social bias in cross-modal text-image retrieval systems, focusing on the interaction between textual queries and image responses. Despite the significant advancements in cross-modal retrieval models, the potential for social bias in their responses remains a pressing concern, necessitating a comprehensive framework for assessment and mitigation. We introduce a novel framework for evaluating social bias in cross-modal retrieval systems, leveraging a new dataset and appropriate metrics specifically designed for this purpose. Our dataset, Social Inclusive Synthetic Professionals Images (SISPI), comprises 49K images generated using state-of-the-art text-to-image models, ensuring a balanced representation of demographic groups across various professional roles. We use this dataset to conduct an extensive analysis of social bias (gender and ethnic) in state of the art cross-modal retrieval deep models, including CLIP, ALIGN, BLIP, FLAVA, COCA, and many others. Using diversity metrics, grounded in the distribution of different demographic groups' images in the retrieval rankings, we provide a quantitative measure of fairness, facilitating a detailed analysis of models' behavior. Our work sheds light on biases present in current cross-modal retrieval systems and emphasizes the importance of training data curation, providing a foundation for future research and development towards more equitable and unbiased models. The dataset and code of our framework is publicly available at https://sispi-benchmark.github.io/sispi-benchmark/."
},
{
"venue": "CHIIR",
"title": "Challenging but satisfying: an exploration of the practices and affective aspects of personal photograph deletion",
"authors": [
"Helene Hellmich",
"Jesse David Dinneen"
],
"year": 2025,
"pdf_url": "https://www.emerald.com/jd/article-pdf/81/7/443/10341174/jd-04-2025-0101en.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1108/jd-04-2025-0101",
"abstract": "Purpose With the pervasive use of smartphone cameras, individuals accumulate increasingly large personal photograph collections which could have a negative impact on individuals and the environment. One way to address collection growth is by deleting, but how people delete is not well understood and thus how to best support them in that action is not clear. Further, prior work suggests such deletion is associated with negative emotions and a lack of enjoyment. Design/methodology/approach This exploratory study used interviews and guided tours with twelve participants to examine deletion practices with personal photographs and the accompanying affective aspects. Findings Deleting, instead of being reduced to the action of placing a photograph into the trash, was characterised as a process over time, requiring a set of interconnected activities. Nine affective aspects were found to play central roles in the deletion of personal photographs but, surprisingly, participants experienced more positive affective aspects, especially after deleting. Originality/value The results provide a novel characterisation of deleting as a process and explore the temporal dimension of its affective aspects. Recommendations are identified for individuals, information professionals and system designers to encourage and improve the deletion of personal photographs, aiming to create more sustainable information practices and improve digital well-being."
},
{
"venue": "CHIIR",
"title": "Exploring Practices, Challenges, and Design Implications for Citation Foraging, Management, and Synthesis",
"authors": [
"Xinrui Fang",
"Anran Xu",
"Sylvain Malacria",
"Koji Yatani"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3706599.3719883",
"source": "openalex",
"doi": "https://doi.org/10.1145/3706599.3719883",
"abstract": "Citations play a crucial role in reinforcing knowledge construction, sparking new ideas, and fostering science communication.However, researchers often encounter difficulties in foraging, managing, and synthesizing citations amid the rapid growth of academic publications.To better understand researchers' practices and challenges in citation related activities, we conducted semi-structured interviews with 12 researchers in the fields of HCI and AI.Our key findings include: (1) researchers are unable to fully track and digest all relevant work; (2) current citation management tools fall short of meeting researchers' needs; (3) \"cherry-picking\" (write the statement first, then search for supporting references to strengthen its credibility and accuracy) is a common practice in citation synthesis; and (4) citation foraging, management, and synthesis workflows are disconnected and lack consistency.Our design implications provide insight into the development of interactive systems that more effectively support researchers in their citation activities."
},
{
"venue": "CHIIR",
"title": "Islamic Values-Based HRM and Its Impact on Student Achievement in Islamic Education",
"authors": [
"Muhamad Faizin",
"Rudi Sulaeman",
"Cepi Budiyanto",
"Ilham Fahmi",
"Muchdjabir Wahid"
],
"year": 2025,
"pdf_url": "https://nidhomulhaq.uacmjk.ac.id/index.php/ndh/article/download/243/95",
"source": "openalex",
"doi": "https://doi.org/10.31538/ndhq.v10i3.243",
"abstract": "This study aims to analyze the implementation of Islamic values-based management of teaching and educational staff and its impact on improving student academic achievement. This study employed a qualitative approach, utilizing a field study design involving 15 informants, comprising study program heads, lecturers, educational staff, and students. Data were collected through in-depth interviews, observations of academic activities, and reviews of performance documents and SOPs for human resource management, then analyzed using the Miles & Huberman interactive model. The results showed that the integration of Islamic values contributed significantly to improving student achievement, with the average GPA increasing from 3.50 to 3.83 and the on-time graduation rate increasing from 78% to 93% in the last four years. Recruitment based on value alignment, competency development that combines pedagogical and ethical dimensions, and performance evaluation with moral and academic indicators has a positive effect on student motivation and achievement. The field findings show that the value of shura fosters participatory leadership and collective accountability; amanah enhances the personal integrity of teaching staff; ihsan promotes a culture of quality and learning innovation; while itqan serves as the basis for work excellence and academic rigor. Analytically, this finding extends the theories of Value-Based Human Resource Management and Transformational Leadership by incorporating an Islamic spiritual dimension, which has been shown to influence organizational behavior and student learning outcomes. The novelty of this research lies in the empirical evidence that the integration of Islamic values into HRM functions at the study program level directly improves student achievement, and it presents a theoretical synthesis between modern educational management practices and Islamic spirituality that has previously been rarely empirically explored in the context of Islamic higher education in Indonesia."
},
{
"venue": "CHIIR",
"title": "Quantification and Evolution of Online Public Opinion Heat Considering Interactive Behavior and Emotional Conflict",
"authors": [
"Zhengyi Sun",
"Deyao Wang",
"Zhaohui Li"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/1099-4300/27/7/701/pdf?version=1751199277",
"source": "openalex",
"doi": "https://doi.org/10.3390/e27070701",
"abstract": "With the rapid development of the Internet, the speed and scope of sudden public events disseminating in cyberspace have grown significantly. Current methods of quantifying public opinion heat often neglect emotion-driven factors and user interaction behaviors, making it difficult to accurately capture fluctuations during dissemination. To address these issues, first, this study addressed the complexity of interaction behaviors by introducing an approach that employs the information gain ratio as a weighting indicator to measure the \"interaction heat\" contributed by different interaction attributes during event evolution. Second, this study built on SnowNLP and expanded textual features to conduct in-depth sentiment mining of large-scale opinion texts, defining the variance of netizens' emotional tendencies as an indicator of emotional fluctuations, thereby capturing \"emotional heat\". We then integrated interactive behavior and emotional conflict assessment to achieve comprehensive heat index to quantification and dynamic evolution analysis of online public opinion heat. Subsequently, we used Hodrick-Prescott filter to separate long-term trends and short-term fluctuations, extract six key quantitative features (number of peaks, time of first peak, maximum amplitude, decay time, peak emotional conflict, and overall duration), and applied K-means clustering algorithm (K-means) to classify events into three propagation patterns, which are extreme burst, normal burst, and long-tail. Finally, this study conducted ablation experiments on critical external intervention nodes to quantify the distinct contribution of each intervention to the propagation trend by observing changes in the model's goodness-of-fit (R2) after removing different interventions. Through an empirical analysis of six representative public opinion events from 2024, this study verified the effectiveness of the proposed framework and uncovered critical characteristics of opinion dissemination, including explosiveness versus persistence, multi-round dissemination with recurring emotional fluctuations, and the interplay of multiple driving factors."
},
{
"venue": "CHIIR",
"title": "Navigating the design of simulated exercising peers: insights from a participatory design study",
"authors": [
"Alessandro Silacci",
"Mauro Cherubini",
"Maurizio Caon"
],
"year": 2025,
"pdf_url": "https://www.frontiersin.org/journals/digital-health/articles/10.3389/fdgth.2025.1551966/pdf",
"source": "openalex",
"doi": "https://doi.org/10.3389/fdgth.2025.1551966",
"abstract": "Background: To fight sedentary lifestyles, researchers have introduced various technological interventions aimed at promoting physical activity through social support. These interventions encourage people to exercise together, maintaining high levels of motivation. However, the unpredictable nature of human peers makes it challenging to control behavior and balance these interventions effectively. Artificial intelligence agents, on the other hand, can provide consistent social support and are more controllable. Hence, we propose Simulated Exercising Peers (SEPs) as a promising solution for providing agent-based social support for physical activity. Method: Participatory design sessions were conducted, involving young adults in the creation of SEP-based interventions. Sixteen participants generated four prototypes that varied in aesthetics, behavior, and communication style, with outcomes analyzed through the lens of Self-Determination Theory to better understand the motivational implications of each design. Results: Findings highlight key components crucial for designing SEPs that enhance acceptance and efficiently integrate into physical activity interventions. Additionally, the study revealed how the aesthetics and behavior of SEPs could potentially deceive users, which can lead to user disengagement from interventions involving SEPs. Participants also defined two distinct social roles for the SEPs, i.e., coach, and companion, each associated with unique communication styles. Conclusion: This study offers five design guidelines for the development of SEPs, AI agents aimed at promoting physical activity through social support, and highlights opportunities for their integration into broader physical activity interventions."
},
{
"venue": "CHIIR",
"title": "Measuring What Matters: Connecting AI Ethics Evaluations to System Attributes, Hazards, and Harms",
"authors": [
"Shalaleh Rismani",
"Renee Shelby",
"Leah Davis",
"Negar Rostamzadeh",
"AJung Moon"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AIES/article/download/36706/38844",
"source": "openalex",
"doi": "https://doi.org/10.1609/aies.v8i3.36706",
"abstract": "Over the past decade, an ecosystem of measures has emerged to evaluate the social and ethical implications of AI systems, largely shaped by high-level ethics principles. These measures are developed and used in fragmented ways, without adequate attention to how they are situated in AI systems. In this paper, we examine how existing measures used in the computing literature map to AI system components, attributes, hazards, and harms. Our analysis draws on a scoping review resulting in nearly 800 measures corresponding to 11 AI ethics principles. We find that most measures focus on four principles – fairness, transparency, privacy, and trust – and primarily assess model or output system components. Few measures account for interactions across system elements, and only a narrow set of hazards is typically considered for each harm type. Many measures are disconnected from where harm is experienced and lack guidance for setting meaningful thresholds. These patterns reveal how current evaluation practices remain fragmented, measuring in pieces rather than capturing how harms emerge across systems. Framing measures with respect to system attributes, hazards, and harms can strengthen regulatory oversight, support actionable practices in industry, and ground future research in systems-level understanding."
},
{
"venue": "CHIIR",
"title": "Evaluating Conversational Recommender Systems via Large Language Models: A User-Centric Framework",
"authors": [
"Nuo Chen",
"Quanyu Dai",
"Xiaoyu Dong",
"Wang, Piaohong",
"Jia, Qinglin",
"Du, Zhaocheng",
"Dong, Zhenhua",
"Wu, Xiao-Ming"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2501.09493",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2501.09493",
"abstract": "Conversational recommender systems (CRSs) integrate both recommendation and dialogue tasks, making their evaluation uniquely challenging. Existing approaches primarily assess CRS performance by separately evaluating item recommendation and dialogue management using rule-based metrics. However, these methods fail to capture the real human experience, and they cannot draw direct conclusions about the system's overall performance. As conversational recommender systems become increasingly vital in e-commerce, social media, and customer support, the ability to evaluate both recommendation accuracy and dialogue management quality using a single metric, thereby authentically reflecting user experience, has become the principal challenge impeding progress in this field. In this work, we propose a user-centric evaluation framework based on large language models (LLMs) for CRSs, namely Conversational Recommendation Evaluator (CoRE). CoRE consists of two main components: (1) LLM-As-Evaluator. Firstly, we comprehensively summarize 12 key factors influencing user experience in CRSs and directly leverage LLM as an evaluator to assign a score to each factor. (2) Multi-Agent Debater. Secondly, we design a multi-agent debate framework with four distinct roles (common user, domain expert, linguist, and HCI expert) to discuss and synthesize the 12 evaluation factors into a unified overall performance score. Furthermore, we apply the proposed framework to evaluate four CRSs on two benchmark datasets. The experimental results show that CoRE aligns well with human evaluation in most of the 12 factors and the overall assessment. Especially, CoRE's overall evaluation scores demonstrate significantly better alignment with human feedback compared to existing rule-based metrics."
},
{
"venue": "CHIIR",
"title": "Exploring Self-dehumanization as a Factor in Misinformation Belief and Spread",
"authors": [
"Andrew Weiss",
"Souvick Ghosh",
"Frances Johnson"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3698204.3716467",
"source": "openalex",
"doi": "https://doi.org/10.1145/3698204.3716467",
"abstract": "The role of dehumanization is little explored in the literature of misinformation studies, primarily seen as a mechanism of rightwing authoritarianism (RWA) to denigrate members of targeted outgroups.One noted aspect of dehumanization is the unexpected impact it has on the person enacting it as well as victims, resulting in the noticeable phenomenon of self-dehumanization, in which persons deny their own humanity.Notably, a link to selfdehumanization in information behavior may be evident in the foundational research of Eflreda Chatman, whose examination of information poverty, life in the round and normative behavioral theory resulted in sometimes perplexing findings where those in marginalized groups often would refuse to seek out information helpful to them.It is theorized that a missing link in Chatman's approach, accounting for some of the unexpected results in her research, may be the effect of self-dehumanization on marginalized groups and their members.The result of such self-dehumanization effects may impact information use, pointing toward another factor in the way that misinformation is believed or spread.This paper shows a novel way forward in the field of information science and information behavior, examining an important aspect of how misinformation may come to be believed and shared within smaller worlds and codified in marginalized groups' normative behaviors.Finally, the paper points the way toward a nascent conceptualization of a 'misinformation need' inherent to the normative behaviors of those within marginalized groups."
},
{
"venue": "CHIIR",
"title": "The Sharply Decreasing Disruptiveness of HCI",
"authors": [
"Zhilong Chen",
"Yong Li"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3706598.3713917",
"source": "openalex",
"doi": "https://doi.org/10.1145/3706598.3713917",
"abstract": "How creative is HCI research?Although creativity has been a notable theme in HCI, the landscape of the creativity of HCI research itself remains unclear.In this paper, we address this by measuring the disruptiveness of HCI research, one important dimension distinguishing the level of creativity, through a large-scale datadriven bibliometric analysis.By quantitatively tracing its evolution over the past 40 years, we find that the disruptiveness of HCI is decreasing sharply, even at a faster speed than the global average across all fields.We characterize the patterns shown by the themes, knowledge use, and authorship of disruptive papers in HCI, and identify how they associate with disruptiveness, e.g., the positive relationship between author freshness and disruptiveness.Based on our results, we discuss practical implications to improve and secure disruptiveness and creativity in HCI."
},
{
"venue": "CHIIR",
"title": "A Comparative Study of Large Language Models and Traditional Privacy Measures to Evaluate Query Obfuscation Approaches",
"authors": [
"Francesco Luigi De Faveri",
"Guglielmo Faggioli",
"Nicola Ferro"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730158",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730158",
"abstract": "When interacting with an Information Retrieval (IR) system, users might disclose personal information, such as medical details, through their queries. Thus, assessing the level of privacy granted to users when querying an IR system is essential to determine the confidentiality of submitted sensitive data. Query obfuscation protocols have traditionally been employed to obscure a user's real information need when retrieving documents. In these protocols, the query is modified employing ε-Differential Privacy (DP) obfuscation mechanisms, which alter query terms according to a predefined privacy budget ε. While this budget ensures formal mathematical guarantees, it provides only limited guarantees of the privacy experienced by the user and calls for empirical privacy evaluation to be carried out. Such privacy assessments employ lexical and semantic similarity measures between the original and obfuscated queries. In this study, we explore the role of Large Language Models (LLMs) in privacy evaluation, simulating a scenario where users employ such models to determine whether their input has been effectively privatized. Our primary research objective is to determine whether LLMs provide a novel perspective on privacy estimation and if their assessments serve as a proxy for traditional similarity metrics, such as the Jaccard and cosine similarity derived from Transformer-based sentence embeddings. Our findings reveal a positive correlation between LLMs-generated privacy scores and cosine similarity computed using different Transformer architectures. This suggests that LLM assessments act as a proxy for similarity-based measures."
},
{
"venue": "CHIIR",
"title": "A Systematic Review of Human-AI Co-Creativity",
"authors": [
"Sahab Singh",
"Koen V. Hindriks",
"Dirk Heylen",
"Kim Baraka"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2506.21333",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2506.21333",
"abstract": "The co creativity community is making significant progress in developing more sophisticated and tailored systems to support and enhance human creativity. Design considerations from prior work can serve as a valuable and efficient foundation for future systems. To support this effort, we conducted a systematic literature review of 62 papers on co-creative systems. These papers cover a diverse range of applications, including visual arts, design, and writing, where the AI acts not just as a tool but as an active collaborator in the creative process. From this review, we identified several key dimensions relevant to system design: phase of the creative process, creative task, proactive behavior of the system, user control, system embodiment, and AI model type. Our findings suggest that systems offering high user control lead to greater satisfaction, trust, and a stronger sense of ownership over creative outcomes. Furthermore, proactive systems, when adaptive and context sensitive, can enhance collaboration. We also extracted 24 design considerations, highlighting the value of encouraging users to externalize their thoughts and of increasing the system's social presence and transparency to foster trust. Despite recent advancements, important gaps remain, such as limited support for early creative phases like problem clarification, and challenges related to user adaptation to AI systems."
},
{
"venue": "CHIIR",
"title": "BALI: Enhancing Biomedical Language Representations through Knowledge Graph and Language Model Alignment",
"authors": [
"Andrey Sakhovskiy",
"Elena Tutubalina"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729901",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729901",
"abstract": "In recent years, there has been substantial progress in using pretrained Language Models (LMs) on a range of tasks aimed at improving the understanding of biomedical texts. Nonetheless, existing biomedical LLMs show limited comprehension of complex, domain-specific concept structures and the factual information encoded in biomedical Knowledge Graphs (KGs). In this work, we propose BALI (Biomedical Knowledge Graph and Language Model Ali gnment), a novel joint LM and KG pre-training method that augments an LM with external knowledge by the simultaneous learning of a dedicated KG encoder and aligning the representations of both the LM and the graph. For a given textual sequence, we link biomedical concept mentions to the Unified Medical Language System (UMLS) KG and utilize local KG subgraphs as cross-modal positive samples for these mentions. Our empirical findings indicate that implementing our method on several leading biomedical LMs, such as PubMedBERT and BioLinkBERT, improves their performance on a range of language understanding tasks and the quality of entity representations, even with minimal pre-training on a small alignment dataset sourced from PubMed scientific abstracts."
},
{
"venue": "CHIIR",
"title": "Scientometric Analysis of the German IR Community within TREC & CLEF",
"authors": [
"Andreas Konstantin Kruff",
"Philipp Schaer"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2502.03065",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2502.03065",
"abstract": "Within this study, the influence of the German Information Retrieval community on the retrieval campaigns Text Retrieval Conference (TREC) and Conference and Labs of the Evaluation Forum (CLEF) between 2000 and 2022 was analyzed based on metadata provided by OpenAlex and further metadata extracted with the GROBID framework from the publication's full texts. The analysis was conducted at the institutional and researcher levels. It was found that the German IR community, both on the author and institution level, mainly contributed to CLEF. Furthermore, it was shown that productivity follows the assumptions made by Lotka's Law."
},
{
"venue": "CHIIR",
"title": "Personalized Knowledge Gain Estimation Through Query-Driven Learning Goal Inference in Search As Learning",
"authors": [
"Hadi Nasser",
"Célia da Costa Pereira",
"Cathy Escazut",
"Andrea G. B. Tettamanzi"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3698204.3716463",
"source": "openalex",
"doi": "https://doi.org/10.1145/3698204.3716463",
"abstract": "The measurement of knowledge gain in information retrieval has garnered significant attention, particularly in evaluating its relationship with user behaviors and its role in enhancing the learning experience by tracking progress toward learning objectives.Previous studies have focused on estimating knowledge acquisition based on a static and predefined representation of learning objectives for each search topic, assessing users' progress toward these fixed goals.However, users often differ in their interests, focusing on various aspects or subtopics within the same broader topic, which they express through diverse, topic-related queries.In this paper, we propose a personalized approach to knowledge gain measurement by adapting and extending an existing method for inferring learning subgoals based on user queries.Our approach extends this method by dynamically determining the number of subgoals for each user query, rather than using a fixed number.Knowledge gain estimation is then conducted based on these individualized subgoals while also incorporating the user's prior knowledge of the search topic to enhance personalization.Using 10 different topics, we compare our approach to a baseline method in which the learning goal representation remains uniform for all users within a given topic."
},
{
"venue": "CHIIR",
"title": "Designing Interactive Multimodal Information Retrieval and Access for Heads Up Computing (DIMIRA-HUC)",
"authors": [
"Haiming Liu",
"Shengdong Zhao",
"Silang Wang",
"Preben Hansen",
"Ian Oakley",
"Khanh-Duy Le"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3698204.3716482",
"source": "openalex",
"doi": "https://doi.org/10.1145/3698204.3716482",
"abstract": "The advancement of wearable intelligent systems presents a unique opportunity to transform how humans interact with digital content. This workshop explores the design of Interactive Multimodal Information Retrieval and Access systems specifically tailored for Heads-Up Computing environments. By leveraging multimodal inputs, such as voice, gaze, and gesture, these systems enable real-time, hands-free access to digital information, facilitating seamless and efficient interaction. The goal is to support tasks requiring rapid information access in dynamic environments while ensuring users remain \"heads-up\" and engaged with the real world. This half-day workshop will share research outcomes and best practices, foster community building, and facilitate discussions on key challenges. By bringing together researchers and practitioners, it aims to drive further advancements in both research and practical applications within this rapidly evolving field."
},
{
"venue": "CHIIR",
"title": "The Enhanced Subgoal Manager: Supporting Complex Learning During Information Seeking for Creative Tasks",
"authors": [
"Kelsey Urgo",
"Yuan Li"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3698061.3734403",
"source": "openalex",
"doi": "https://doi.org/10.1145/3698061.3734403",
"abstract": "Traditional search systems and generative AI (GenAI) tools are designed for quick information retrieval, yet creative tasks involving complex learning require sustained engagement and deep cognitive processing.Existing AI and search tools rarely provide the structured support necessary for navigating the iterative, exploratory, and reflective nature of creative tasks.We introduce the Enhanced Subgoal Manager (EnSM), a tool that integrates GenAI that is specifically designed to scaffold complex learning during creative information-seeking tasks.Through structured reflection, goal-setting, and iterative adaptation across multiple sessions, the EnSM supports learners engaging in complex cognitive processes such as understanding abstract concepts, critically evaluating information, and synthesizing multiple sources.This work contributes to a broader reconceptualization of AI-assisted learning from rapid information access toward sustained, meaningful engagement."
},
{
"venue": "CHIIR",
"title": "Classifying Term Variants in Query Formulation",
"authors": [
"Nuha Abu Onq",
"Mark Sanderson",
"Falk Scholer"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729924",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729924",
"abstract": "Formulating queries is a challenging stage of the search process. This study investigates how crowd workers formulate an initial query for a common information need described in a backstory, resulting in diverse query variations. Using the UQV100 dataset of information need backstories and corresponding queries, we analyze the variations. Our findings show that 70% of the query terms used in crowd worker queries did not appear in the backstory text. Examining such terms we developed a taxonomy of search strategies, with the most common being semantic variations of backstory terms, followed by information type specifications. Additionally, we categorized the backstories by cognitive complexity, showing that higher complexity led to greater diversity in query variations and a wider range of term variant categories. This study highlights the importance of accounting for query variations, term variants, user strategies, and cognitive complexity in designing search systems and test collections to better align with users' information needs, influenced by the cognitive demands of a task, and enhance system performance and usability."
},
{
"venue": "CHIIR",
"title": "RMIT-ADM+S at the SIGIR 2025 LiveRAG Challenge",
"authors": [
"Kun Ran",
"Shuoqi Sun",
"Khoi Nguyen Dinh Anh",
"Damiano Spina",
"Oleg Zendel"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2506.14516",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2506.14516",
"abstract": "This paper presents the RMIT--ADM+S winning system in the SIGIR 2025 LiveRAG Challenge. Our Generation-Retrieval-Augmented Generation (G-RAG) approach generates a hypothetical answer that is used during the retrieval phase, alongside the original question. G-RAG also incorporates a pointwise large language model (LLM)-based re-ranking step prior to final answer generation. We describe the system architecture and the rationale behind our design choices. In particular, a systematic evaluation using the Grid of Points approach and N-way ANOVA enabled a controlled comparison of multiple configurations, including query variant generation, question decomposition, rank fusion strategies, and prompting techniques for answer generation. The submitted system achieved the highest Borda score based on the aggregation of Coverage, Relatedness, and Quality scores from manual evaluations, ranking first in the SIGIR 2025 LiveRAG Challenge."
},
{
"venue": "CHIIR",
"title": "ISMIE: A Framework to Characterize Information Seeking in Modern Information Environments",
"authors": [
"Shuoqi Sun",
"Danula Hettiachchi",
"Damiano Spina"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3767695.3769509",
"source": "openalex",
"doi": "https://doi.org/10.1145/3767695.3769509",
"abstract": "The modern information environment (MIE) is increasingly complex, shaped by a wide range of techniques designed to satisfy users' information needs. Information seeking (IS) models are effective mechanisms for characterizing user-system interactions. However, conceptualizing a model that fully captures the MIE landscape poses a challenge. We argue: Does such a model exist? To address this, we propose the Information Seeking in Modern Information Environments (ISMIE) framework as a fundamental step. ISMIE conceptualizes the information seeking process (ISP) via three key concepts: Components (e.g., Information Seeker), Intervening Variables (e.g., Interactive Variables), and Activities (e.g., Acquiring). Using ISMIE's concepts and employing a case study based on a common scenario - misinformation dissemination – we analyze six existing IS and information retrieval (IR) models to illustrate their limitations and the necessity of ISMIE. We then show how ISMIE serves as an actionable framework for both characterization and experimental design. We characterize three pressing issues and then outline two research blueprints: a user-centric, industry-driven experimental design for the authenticity and trust crisis to AI-generated content and a system-oriented, academic-driven design for tackling dopamine-driven content consumption. Our framework offers a foundation for developing IS and IR models to advance knowledge on understanding human interactions and system design in MIEs."
},
{
"venue": "CHIIR",
"title": "“I'm Looking for a Book with a Big Watermelon and Many Small Ants”: An Exploratory Study on a Visualized Search System for Preschoolers' Picture Book Search",
"authors": [
"Jinghan Zhang",
"Yi Xie",
"Junyang Ma",
"Zhihan Zheng",
"Shenkang Zhai",
"Pianran Wang"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3698204.3716469",
"source": "openalex",
"doi": "https://doi.org/10.1145/3698204.3716469",
"abstract": "This paper introduces the development process of a preschool children oriented visualized picture book search system—a picture book search system designed specifically for children and their parents. This study involves a thorough literature review and a field investigation to gain insights into the actual needs of children. Besides, it provides an efficient picture book search service by designing a child-friendly interface and integrating advanced search technology. The study employs heuristic evaluation and multiple performance indicators to ensure the system's high recall, precision and user experience. The system helps to promote the cultivation of children's reading interest and the improvement of reading ability, and will be implemented in libraries."
},
{
"venue": "CHIIR",
"title": "Understanding Mental Models of Generative Conversational Search and The Effect of Interface Transparency",
"authors": [
"Chadha Degachi",
"Samuel Kernan Freire",
"Evangelos Niforatos",
"Gerd Kortuem"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2506.03807",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2506.03807",
"abstract": "The experience and adoption of conversational search is tied to the accuracy and completeness of users' mental models -- their internal frameworks for understanding and predicting system behaviour. Thus, understanding these models can reveal areas for design interventions. Transparency is one such intervention which can improve system interpretability and enable mental model alignment. While past research has explored mental models of search engines, those of generative conversational search remain underexplored, even while the popularity of these systems soars. To address this, we conducted a study with 16 participants, who performed 4 search tasks using 4 conversational interfaces of varying transparency levels. Our analysis revealed that most user mental models were too abstract to support users in explaining individual search instances. These results suggest that 1) mental models may pose a barrier to appropriate trust in conversational search, and 2) hybrid web-conversational search is a promising novel direction for future search interface design."
},
{
"venue": "CHIIR",
"title": "Creating Coherence in Federated Non-Negative Matrix Factorization",
"authors": [
"Sebastian Dalleiger",
"Aristides Gionis"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/33772/35927",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i15.33772",
"abstract": "In many real-world applications, data is inherently decentralized, necessitating data analysis methods that prioritize privacy while delivering interpretable results. Federated Non-Negative Matrix Factorization (FedNMF) meets this requirement by factorizing latent components from distributed data that cannot be freely shared among clients. A significant challenge in FedNMF arises when clients converge on different solutions due to prolonged independent optimization, leading to drift and incoherent models. While Federated Learning (FL) typically mitigates drift through frequent synchronizations and strong regularization, it often overlooks critical properties of Non-Negative Matrix Factorization, such as permutation invariance. As a result, solutions from FedNMF clients may be misidentified by FL drift as distinct, despite being equivalent. Using an alignment-aware drift, we create coherence through proximal optimization and barycenter aggregation for FedNMF. We analyze the computational complexity of our approach, provide efficient heuristics, and ensure the convergence of our algorithms. On a diverse set of real-world and synthetic datasets, we demonstrate the effectiveness of our methods."
},
{
"venue": "CHIIR",
"title": "Multi-stage Large Language Model Pipelines Can Outperform GPT-4o in Relevance Assessment",
"authors": [
"Julian A. Schnabel",
"Johanne R. Trippas",
"Falk Scholer",
"Danula Hettiachchi"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2501.14296",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2501.14296",
"abstract": "The effectiveness of search systems is evaluated using relevance labels that indicate the usefulness of documents for specific queries and users. While obtaining these relevance labels from real users is ideal, scaling such data collection is challenging. Consequently, third-party annotators are employed, but their inconsistent accuracy demands costly auditing, training, and monitoring. We propose an LLM-based modular classification pipeline that divides the relevance assessment task into multiple stages, each utilising different prompts and models of varying sizes and capabilities. Applied to TREC Deep Learning (TREC-DL), one of our approaches showed an 18.4% Krippendorff's $α$ accuracy increase over OpenAI's GPT-4o mini while maintaining a cost of about 0.2 USD per million input tokens, offering a more efficient and scalable solution for relevance assessment. This approach beats the baseline performance of GPT-4o (5 USD). With a pipeline approach, even the accuracy of the GPT-4o flagship model, measured in $α$, could be improved by 9.7%."
},
{
"venue": "CHIIR",
"title": "Sistem Rekomendasi Musik Spotify Berdasarkan Listening History Pengguna",
"authors": [
"Ichsan Madani",
"Hasrullah"
],
"year": 2025,
"pdf_url": "https://jig.rivierapublishing.id/index.php/rv/article/download/276/505",
"source": "openalex",
"doi": "https://doi.org/10.58344/jig.v3i2.276",
"abstract": "Dengan meningkatnya popularitas musik digital, pengguna sering kali menemukan kesulitan menemukan musik yang sesuai dengan preferensi dan konteks mereka. Sistem rekomendasi yang ada saat ini kurang personal, memenuhi kebutuhan pendengar. Tujuan penelitian ini adalah untuk menciptakan dan menguji sistem rekomendasi musik yang bergantung pada riwayat mendengarkan yang digunakan pengguna di platform Spotify. Untuk memberikan saran yang lebih personal dan relevan untuk penelitian ini, kami menggunakan pendekatan hybrid yang menggabungkan metode filtrasi kolaboratif dan filtrasi berbasis konten. Informasi seperti artis, genre, lagu, dan waktu pengguna diambil dari rekaman mendengarkan. Pemodelan sistem rekomendasi, preprocessing, dan pengumpulan data adalah bagian dari metodologi penelitian. Dalam hal akurasi dalam merekomendasikan musik, hasil evaluasi sistem menunjukkan tingkat presisi sebesar 87%, recall sebesar 82%, dan Mean Absolute Error (MAE) sebesar 0,15. Temuan ini menunjukkan bahwa sistem rekomendasi yang dikembangkan mampu meningkatkan kepuasan pengguna dengan memberikan rekomendasi musik yang relevan berdasarkan preferensi dan konteks individu, termasuk aktivitas olahraga di sore hari dalam cuaca cerah. Penelitian ini diharapkan dapat menjadi dasar untuk pengembangan lebih lanjut dalam rekomendasi musik yang lebih kompleks dan dinamis. Ini juga telah memberikan kontribusi signifikan dalam pengembangan teknologi rekomendasi personal untuk layanan streaming musik."
},
{
"venue": "CHIIR",
"title": "From Seo To Answer Engine Optimization (aeo): Generative Ai and the Transformation of Search Visibility",
"authors": [
"HİDAYET KARAMUK"
],
"year": 2025,
"pdf_url": "https://avesis.kocaeli.edu.tr/publication/details/d484d872-4228-4b17-b048-56365f8c55bf/oai/document.pdf",
"source": "openalex",
"doi": "https://doi.org/10.70269/10.70269/7348479369",
"abstract": "This book chapter examines the transformation of search visibility from traditional Search Engine Optimization (SEO) to Answer Engine Optimization (AEO) in the context of generative artificial intelligence and large language models. As AI-driven search engines increasingly provide direct, synthesized answers rather than ranked links, the logic of digital visibility, content production, and brand discoverability is undergoing a fundamental shift. The chapter conceptually analyzes how generative AI reshapes information retrieval, user search behavior, and the criteria through which content gains prominence in answer-based search environments. Building on this transformation, the study discusses the limitations of keyword-centered SEO practices and highlights the growing importance of semantic relevance, user intent, structured data, topical authority, and trust signals in AI-mediated search ecosystems. The chapter further explores how brands and content creators must adapt their digital marketing strategies to align with answer engines that prioritize contextual understanding, credibility, and usefulness over traditional ranking metrics. By integrating insights from digital marketing, information systems, and AI studies, this chapter offers a comprehensive framework for understanding AEO as an emerging paradigm of search visibility. The study aims to contribute to the academic literature while providing strategic implications for marketers navigating AI-driven search environments."
},
{
"venue": "CHIIR",
"title": "ChatR1: Reinforcement Learning for Conversational Reasoning and Retrieval Augmented Question Answering",
"authors": [
"Simon Lupart",
"Mohammad Aliannejadi",
"Evangelos Kanoulas"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2510.13312",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2510.13312",
"abstract": "We present ChatR1, a reasoning framework based on reinforcement learning (RL) for conversational question answering (CQA). Reasoning plays an important role in CQA, where user intent evolves across dialogue turns, and utterances are often underspecified, requiring contextual interpretation, query reformulation, and dynamic coordination between retrieval and generation. Unlike static `rewrite, retrieve, and generate' pipelines, ChatR1 interleaves search and reasoning across turns, enabling exploratory and adaptive behaviors learned through RL. To address the challenge of sparse and delayed rewards in RL, we propose an intent-aware reward that provides turn-level feedback by aligning retrieval and reasoning with evolving user goals. ChatR1 demonstrates strong performance on both 3B and 7B model backbones, outperforming competitive models on five CQA datasets, measured by different metrics (F1, BERTScore, and LLM-as-judge). We include a diverse set of CQA datasets to cover topic shifts, evolving intents, mixed-initiative dialogues, and multi-document grounding, testing ChatR1's performance from various aspects. Ablation studies confirm the effectiveness of the intent-aware reward. Our analyses further reveal diverse reasoning trajectories and effective use of the search tool. ChatR1 also generalizes robustly across domains, demonstrating that RL-based reasoning enables more flexible and context-aware behavior than static CQA pipelines."
},
{
"venue": "CHIIR",
"title": "An Analysis of the Innovative Development Path of the E-Sports Industry: Taking TJ Sports as an Example",
"authors": [
"Li Weng"
],
"year": 2025,
"pdf_url": "https://jehss.com/index.php/ojs/article/download/8/7",
"source": "openalex",
"doi": "https://doi.org/10.54097/th7qga78",
"abstract": "With the rapid development of internet technology and the booming digital entertainment industry, e-sports have emerged as a new form of competition, quickly rising to prominence on a global scale and becoming a focal point of attention. This study aims to explore the innovative development path of China's e-sports industry, with a deep analysis of TJ Sports as a case study. By examining the development history, platform attributes, construction of the e-sports ecosystem, and innovative strategies of TJ Sports, the key factors behind its success are revealed. The research indicates that TJ Sports has achieved rapid development and a leading market position by continuously optimizing its online platform, constructing a comprehensive e-sports ecosystem, and implementing innovative strategies. The study finds that China's e-sports industry faces several challenges, including policy regulation, market competition, and talent shortages. These issues require collective efforts to address in order to promote the industry's sustained and healthy development."
},
{
"venue": "CHIIR",
"title": "Advancing Chichewa IR",
"authors": [
"Stanley Ndebvu",
"Reuben Moyo",
"Catherine Chavula"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730272",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730272",
"abstract": "Malawi is home to over ten local languages, including Chichewa, yet many of these languages lack both printed and digital resources. Consequentially, access to information in these languages is limited, and this hinders knowledge sharing, which may potentially impact socio-economic development. In this paper, we discuss our work on developing language resources and tools for Chichewa. We begin by providing an overview of the Chichewa language, and highlight its inherent complexities that require new approaches to informa tion retrieval (IR) and natural language processing (NLP). We then present our past, current, and ongoing research and conclude with future directions. Our goal is to engage with the IR community to discuss how we can advance IR for low resource languages (LRLs) like Chichewa."
},
{
"venue": "CHIIR",
"title": "RMUTTOBot: Transforming University Admission Services with a TAG-based RAG LLM Chatbot",
"authors": [
"Vipa Thananont",
"Saowakhon Nookhao"
],
"year": 2025,
"pdf_url": "https://li01.tci-thaijo.org/index.php/rmutsvrj/article/download/267581/181384",
"source": "openalex",
"doi": "https://doi.org/10.65411/rst.2026.267581",
"abstract": "Advancements in artificial intelligence, particularly in large language models (LLMs) and retrieval-augmented generation (RAG) techniques, have improved chatbot capabilities for more natural and domain-specific interactions. However, conventional RAG systems, which retrieve information from unstructured text sources like websites and PDFs, exhibit critical failures when applied to the dynamic and precise nature of university information. This research addresses these gaps through the design and development of RMUTTOBot, a domain-specific chatbot providing admissions support for prospective students at Rajamangala University of Technology Tawan-ok (RMUTTO). We propose a novel, lightweight table-augmented generation (TAG) approach that combines a curated, updatable knowledge base for general information with live database queries for real-time, dynamic data. Performance was evaluated using both automated metrics and human assessments across six criteria: semantic similarity, retrieval effectiveness, relevance, fluency, coverage, and consistency. Experimental results show that the TAG-based RAG system significantly outperformed both the baseline LLM-only configuration and PDF-based RAG system, achieving a 12.76% higher BERTF1 score compared to a PDF-based RAG. Human evaluation confirmed the system’s high response relevance and linguistic fluency, with strong inter-rater reliability (Krippendorff’s α > 0.835). These findings demonstrate that combining structured data augmentation with RAG substantially enhances chatbot accuracy, contextual grounding, and completeness, offering a robust framework for intelligent conversational systems in academic domains. The source code and implementation details are publicly available at https://github.com/vipa-thananant/RMUTTOBot."
},
{
"venue": "CHIIR",
"title": "Query Smarter, Trust Better? Exploring Search Behaviours for Verifying News Accuracy",
"authors": [
"David Elsweiler",
"Samy Ateia",
"Markus Bink",
"Gregor Donabauer",
"Marcos Fernández-Pichel",
"Alexander Frummet",
"Udo Kruschwitz",
"David E. Losada",
"Bernd Ludwig",
"Selina Meyer",
"Noel Pascual-Presa"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2504.05146",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2504.05146",
"abstract": "While it is often assumed that searching for information to evaluate misinformation will help identify false claims, recent work suggests that search behaviours can instead reinforce belief in misleading news, particularly when users generate queries using vocabulary from the source articles. Our research explores how different query generation strategies affect news verification and whether the way people search influences the accuracy of their information evaluation. A mixed-methods approach was used, consisting of three parts: (1) an analysis of existing data to understand how search behaviour influences trust in fake news, (2) a simulation of query generation strategies using a Large Language Model (LLM) to assess the impact of different query formulations on search result quality, and (3) a user study to examine how 'Boost' interventions in interface design can guide users to adopt more effective query strategies. The results show that search behaviour significantly affects trust in news, with successful searches involving multiple queries and yielding higher-quality results. Queries inspired by different parts of a news article produced search results of varying quality, and weak initial queries improved when reformulated using full SERP information. Although 'Boost' interventions had limited impact, the study suggests that interface design encouraging users to thoroughly review search results can enhance query formulation. This study highlights the importance of query strategies in evaluating news and proposes that interface design can play a key role in promoting more effective search practices, serving as one component of a broader set of interventions to combat misinformation."
},
{
"venue": "CHIIR",
"title": "URAG: Implementing a Unified Hybrid RAG for Precise Answers in University Admission Chatbots -- A Case Study at HCMUT",
"authors": [
"Long D. Nguyen",
"Tho Quan"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2501.16276",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2501.16276",
"abstract": "With the rapid advancement of Artificial Intelligence, particularly in Natural Language Processing, Large Language Models (LLMs) have become pivotal in educational question-answering systems, especially university admission chatbots. Concepts such as Retrieval-Augmented Generation (RAG) and other advanced techniques have been developed to enhance these systems by integrating specific university data, enabling LLMs to provide informed responses on admissions and academic counseling. However, these enhanced RAG techniques often involve high operational costs and require the training of complex, specialized modules, which poses challenges for practical deployment. Additionally, in the educational context, it is crucial to provide accurate answers to prevent misinformation, a task that LLM-based systems find challenging without appropriate strategies and methods. In this paper, we introduce the Unified RAG (URAG) Framework, a hybrid approach that significantly improves the accuracy of responses, particularly for critical queries. Experimental results demonstrate that URAG enhances our in-house, lightweight model to perform comparably to state-of-the-art commercial models. Moreover, to validate its practical applicability, we conducted a case study at our educational institution, which received positive feedback and acclaim. This study not only proves the effectiveness of URAG but also highlights its feasibility for real-world implementation in educational settings."
},
{
"venue": "CHIIR",
"title": "Operationalizing Knowledge Gain: Implementing and Testing the DKG Metric in Search Environments",
"authors": [
"Rafael Tavares da Silva",
"Sean Wolfgand Matsui Siqueira",
"Marcelo Tibau"
],
"year": 2025,
"pdf_url": "https://sol.sbc.org.br/index.php/webmedia_estendido/article/download/38192/37967",
"source": "openalex",
"doi": "https://doi.org/10.5753/webmedia_estendido.2025.16337",
"abstract": "Searching the Web is increasingly recognized as a process of knowledge construction rather than simple information retrieval, a perspective framed by the paradigm of Searching as Learning (SaL). A central challenge in this domain lies in evaluating the extent to which users actually acquire knowledge during search. Traditional approaches either rely on behavioral proxies, scalable but limited in capturing conceptual change, or structured assessments, which provide direct evidence but are often intrusive. The Degree of Knowledge Gain (DKG) metric addresses this gap by modeling reductions in uncertainty through Shannon’s entropy and integrating semantic similarity between queries and clicked documents. This paper reports on the operationalization of DKG within the CNPq project 3C-BPA: Comportamento de busca, Complexidade da informação e pensamento Crítico na Busca como um Processo de Aprendizagem. Two artifacts were developed: an initial search engine prototype embedding DKG computation, and a Chrome extension that estimates DKG in real time while users employ their preferred search engines. The latter artifact overcame earlier limitations by improving ecological validity, reducing costs, and enabling more natural experimentation. An experiment combined pre- and post-tests, the Concurrent Think-Aloud (CTA) protocol, and the plug-in’s automated logging. Preliminary findings show that DKG values are sensitive to differences in search strategies, with systematic reformulation and evaluation aligning with greater knowledge gains, while disorientation behaviors corresponded to more modest outcomes. A distinctive feature of this study was the active role of an undergraduate researcher, who contributed to artifact development, experiment setup, participant support, transcription, and ongoing content analysis."
},
{
"venue": "CHIIR",
"title": "DispatchQA: A Benchmark for Small Function Calling Language Models in E-Commerce Applications",
"authors": [
"Joachim Daiber",
"Victor Maricato",
"Ayan Sinha",
"Andrew Rabinovich"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.emnlp-industry.154.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.emnlp-industry.154",
"abstract": ""
},
{
"venue": "CHIIR",
"title": "Escaping the Delphic Trap: Providing Variation Affordances to Foster Agency and Resilience in AI-Mediated Sensemaking",
"authors": [
"Lei, Nina"
],
"year": 2025,
"pdf_url": "https://dash.harvard.edu/bitstreams/9c99bd93-3852-48b7-a67b-2643b34e0d9e/download",
"source": "openalex",
"doi": "",
"abstract": "With emergent capabilities of generative AI, many systems have been eager to adopt features providing \\textit{synthesis affordances}: properties that enable and entice users in engaging with AI-synthesized information. There are contexts where offering synthesis affordances may be appropriate---perhaps, even useful. However, approaches taken by popular design patterns in providing them are often not \\textit{resilient}. They fail upon breakdowns stemming from AI limitations, human factors, or information quality issues. More concerningly, these design patterns distribute agency inappropriately between humans and AI by granting excessive agency to AI systems. Such configurations lure users into what we term the ``Delphic Trap,'' where users are enticed to satisfice to suboptimal information practice. Drawing on theories from cognitive and ecological psychology and work in Human-Computer Interaction, we conceptualize \\textit{variation affordances}, defining them as system properties that invite users to engage with the inherent variation within information collections. Systems can offer these affordances through steerable controls that support and invite users in engaging in productive friction during information seeking and sensemaking. We argue that a way to instantiate a more appropriate delegation of agency between humans and AI systems, in this context, is through providing variation affordances; doing so may help users engage in more intentional information actions. We design and evaluate several ways to provide these affordances. Chapter 1 establishes the motivating background for this work. Chapter 2 reviews structure-mapping theory literature to inform approaches for helping users utilize variation. Chapter 3 presents an eye-tracking ablation study (n=24) examining how participants interact with different feature sets providing various variation affordances, discussing functional aspects to consider when implementing these affordances. Chapter 4 examines how common design patterns offering synthesis affordances lack resilience and may cause users to fall into the Delphic Trap. Addressing these risks, we propose a design intervention providing variation affordances through ``AI Highlighters''. Chapter 5 presents a formative study (n=24) assessing user interactions with our design intervention and its variation affordances."
},
{
"venue": "CHIIR",
"title": "DIGITAL RITUALS AND EVERYDAY NARRATIVES: A VISUAL ANTHROPOLOGY OF SHORT-FORM VIDEO CULTURE",
"authors": [
"Ivana Ercegovac",
"Romana Srncova",
"Fajar Mohamed Alhanaee"
],
"year": 2025,
"pdf_url": "https://filolog.rs.ba/index.php/filolog/article/download/663/394",
"source": "openalex",
"doi": "https://doi.org/10.21618/fil2532541e",
"abstract": "This paper examines how short-form video platforms transform everyday routines into recognizable narrative and visual patterns. Through qualitative content analysis of 21 videos from TikTok, Instagram Reels, and YouTube Shorts, this pilot study identifies how compressed sequences of action become procedural micro-narratives. These videos rely on minimal cues of setup, shift, and closure, reinforced by visual grammars and culturally coded symbols that ensure legibility at feed speed. The analysis shows how repetition and template use function as digital rituals that both reflect and standardize everyday storytelling. By combining narratology, semiotics, and digital anthropology, the study contributes to understanding how short-form platforms not only distribute content but also actively shape cultural expression and identity. Methodologically, the adoption of the micro-narrative as a unit of analysis enables comparison across languages and platforms, while also highlighting the influence of algorithmic curation on sample formation. The findings open paths for future research using larger datasets and mixed methods, offering insights into how platforms shape the forms and rhythms of everyday narratives over time."
},
{
"venue": "CHIIR",
"title": "Akıllı Asistan Müşteri Deneyiminin Tekrar Kullanım Niyetine Etkisinde Antropomorfizmin Aracılık Rolü: ChatGPT Örneği",
"authors": [
"Serap Battal",
"Nurettin Ozan BAKIR"
],
"year": 2025,
"pdf_url": "https://dergipark.org.tr/tr/download/article-file/5450266",
"source": "openalex",
"doi": "https://doi.org/10.54439/gupayad.1830069",
"abstract": "Amaç: Yapay zekanın günlük yaşama hızla entegre olmasıyla birlikte, tüketicilerin akıllı asistanlarla kurdukları etkileşim müşteri deneyiminin önemli bir boyutu hâline gelmiştir. Bu bağlamda, insan benzeri özellikler taşıyan yapay zekâ destekli asistanların kullanıcı algısını ve tekrar kullanım niyetini nasıl şekillendirdiğini anlamak, pazarlama literatüründe giderek daha kritik bir araştırma alanı olarak öne çıkmaktadır. Bu çalışmanın amacı, akıllı asistan kullanımında tüketici deneyiminin hem algılanan antropomorfizmin hem de tekrar kullanım niyetine etkilerini incelemek, aynı zamanda algılanan antropomorfizmin tekrar kullanım niyetine etkisini araştırmak, ayrıca algılanan antropomorfizmin tüketici deneyimi ile tekrar kullanım niyeti arasındaki aracılık rolünün olup olmadığını tespit etmektir. Gereç ve Yöntem: Araştırmanın ana kütlesi, Türkiye’de yaşayan ve ChatGPT’yi en az bir kez kullanmış 18 yaş üstü kişilerden oluşmaktadır. Bu araştırmada kolayda örnekleme yöntemi tercih edilmiştir. Veriler, çevrimiçi anket yöntemi ile toplanmıştır. Toplamda 424 anket yapısal eşitlik modellemeleri kullanılarak veri analizine tabi tutulmuştur. Bulgular: Yapılan faktör analizlerine göre, tüketici deneyimi dört faktör, algılanan antropomorfizm ve tekrar kullanım niyeti değişkenleri tek faktör altında ele alınmıştır. Yapılan analizlere göre, tüketici deneyiminin tekrar kullanım niyetini ve algılanan antropomorfizmi etkilediği sonucuna ulaşılmıştır. Ayrıca algılanan antropomorfizmin tekrar kullanım niyeti üzerinde etkisi olduğu anlaşılmıştır. Ancak algılanan antropomorfizmin herhangi bir aracılık etkisine rastlanmamıştır. Bulgular doğrultusunda, antropomorfizmin kullanıcı deneyimi esnasında algılarını yönlendirme potansiyeline sahip olduğu söylenebilir. Sonuç: Çalışmadan elde edilen çıktılar, pazarlamacıların antropomorfik etkileşimleri daha dikkatli ve bilinçli şekilde değerlendirerek markaları destekleyebilecek stratejiler oluşturabileceklerini gösterebilir."
},
{
"venue": "CHIIR",
"title": "Exploring the Impact of Warnings on User Perception towards AI-Generated Content in Search Results",
"authors": [
"Pia Donabauer",
"David Elsweiler"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3746252.3761089",
"source": "openalex",
"doi": "https://doi.org/10.1145/3746252.3761089",
"abstract": "Generative-AI answer boxes have become a default element of modern search engines, yet their responses are not always trustworthy. We study whether a simple disclosure can temper the influence of these answers on users' beliefs and behaviour. In a between-subjects online experiment (N=57) participants formed opinions on one of three controversial topics while interacting with a SERP whose featured answer was produced by ChatGPT. Each participant had a 50% chance of seeing a banner that (i) disclosed the answer's AI origin, (ii) listed three key limitations, and (iii) linked to additional details."
},
{
"venue": "CHIIR",
"title": "TeamCMU at Touché: Adversarial Co-Evolution for Advertisement Integration and Detection in Conversational Search",
"authors": [
"To Eun Kim",
"João Paulo Coelho",
"Gbemileke Onilude",
"Jai Govind Singh"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2507.00509",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2507.00509",
"abstract": "As conversational search engines increasingly adopt generation-based paradigms powered by Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG), the integration of advertisements into generated responses presents both commercial opportunities and challenges for user experience. Unlike traditional search, where advertisements are clearly delineated, generative systems blur the boundary between informational content and promotional material, raising concerns around transparency and trust. In this work, we propose a modular pipeline for advertisement management in RAG-based conversational systems, consisting of an ad-rewriter for seamless ad integration and a robust ad-classifier for detection. We leverage synthetic data to train high-performing classifiers, which are then used to guide two complementary ad-integration strategies: supervised fine-tuning of the ad-rewriter and a best-of-N sampling approach that selects the least detectable ad-integrated response among multiple candidates. Our evaluation focuses on two core questions: the effectiveness of ad classifiers in detecting diverse ad integration strategies, and the training methods that best support coherent, minimally intrusive ad insertion. Experimental results show that our ad-classifier, trained on synthetic advertisement data inspired by marketing strategies and enhanced through curriculum learning, achieves robust detection performance. Additionally, we demonstrate that classifier-guided optimization, through both fine-tuning and best-of-N sampling, significantly improves ad stealth, enabling more seamless integration. These findings contribute an adversarial co-evolution framework for developing more sophisticated ad-aware generative search systems and robust ad classifiers."
},
{
"venue": "CHIIR",
"title": "Analyzing the Naming Conventions of Life Science Data Resources to Inform Human and Computational Findability",
"authors": [
"H.J. Imker",
"Hua Ou"
],
"year": 2025,
"pdf_url": "https://www.biorxiv.org/content/biorxiv/early/2025/10/04/2025.10.02.680112.full.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1101/2025.10.02.680112",
"abstract": "Abstract This study aimed to evaluate the names of life science data resources and consider the impacts on findability, a core feature of the FAIR (Findability, Accessibility, Interoperability, and Reusability) Principles. Utilizing a previously published list of unique data resources, we identified and validated data resources with both common and full names available (n = 1153). From this set, we analyzed characteristics of resource names to identify if any naming conventions have emerged organically. Additionally, since common names are often used in the absence of a resource’s full name, we performed a test to evaluate our ability to infer any meaning from common names. Our results highlight suboptimal naming practices and a wide-spread opaqueness in common names, which poses challenges to resource identification and retrieval by both human-and computationally-centric methods. These results are informative for those who establish and promote data resources as well as for those who search for data to use in individual research projects, develop data discovery systems, analyze the scientific literature, or assess research infrastructure. The findings underscore the value of findability in the FAIR Principles and the current efforts to develop infrastructure that supports more efficient communication and global connectedness."
},
{
"venue": "CHIIR",
"title": "Exploring the Lived Experiences of Designing Digital Storeytelling Among EFL Pre-Service Teachers: A Focus on Creative Thinking",
"authors": [
"Arining Indrasari Ahdi",
"Nur Arifah Drajatı",
"Kristian Adi Putra"
],
"year": 2025,
"pdf_url": "https://doi.org/10.56040/ahdp2217",
"source": "openalex",
"doi": "https://doi.org/10.56040/ahdp2217",
"abstract": "Creativity, one of the crucial skills in the 21st century, has become an increasing focus in language teaching as education has transformed to digital learning. There is an urgent need for educators to incorporate activities such as digital storytelling (DST) to encourage students to explore diverse perspectives and develop innovative solutions. In this study, 35 pre-service teachers (PSTs) of English as a Foreign Language (EFL) were voluntarily recruited in adherence to ethical protocols. We explored the PSTs’ experiences in using their creative thinking skills to create digital storytelling. Data were collected through narrative frames, interviews, and artifacts (digital stories). The findings of this study showed that the participants generated numerous ideas for their stories. They were flexible regarding different perspectives and feedback. Furthermore, they ensured that their stories and media components were original. They also enriched their stories with many components (translation, audio, and others). This indicates that the four components of creative thinking skills (fluency, flexibility, originality, and elaboration) were evident in the PSTs’ experiences. Therefore, creative learning activities need to be incorporated into teacher education programs so that teachers will be familiar with creative learning and can integrate it into their classrooms in the future."
},
{
"venue": "CHIIR",
"title": "A diary study of information‐intensive work tasks in the modern workplace: Investigating task descriptions and task processes",
"authors": [
"Afeng Wang",
"Yiming Zhao",
"Feicheng Ma"
],
"year": 2025,
"pdf_url": "https://onlinelibrary.wiley.com/doi/pdfdirect/10.1002/asi.70037",
"source": "openalex",
"doi": "https://doi.org/10.1002/asi.70037",
"abstract": "Abstract The work‐task‐based framework offers a cohesive perspective for understanding workplace information behavior, guiding empirical exploration of information engagement in modern work environments. This study investigates both task descriptions and task processes of information‐intensive work tasks through diaries and follow‐up interviews to capture authentic user experiences. Data from 52 work tasks across diverse organizations reveal that the most frequent topics include Reference , Business , Science , Society , and Computers , with Intellectual and Decision/Solution product types being predominant. Performers typically begin with moderate or high work task knowledge. On average, each work task involves 2.5 seeking tasks and 5.3 search tasks. Seeking tasks are mainly linked to resolution‐oriented information use, while search tasks rely on external sources for factual resolution and verification. Work task topics, product, prior knowledge, subtasks, and duration significantly influence source selection and information use. As work tasks progress, the number of search tasks and clarification use decreases, whereas resolution and verification use increase. These findings refine theoretical models of task‐driven information behavior and provide practical insights for designing adaptive information systems and AI tools to better support evolving work task processes and enhance work performance."
},
{
"venue": "CHIIR",
"title": "The Role of Digital Archival Governance",
"authors": [
"Aris Baharuddin",
"Afzal Sayed Munna",
"Asmilah Abdullah",
"Asmar Asmar",
"Rudi Salam"
],
"year": 2025,
"pdf_url": "https://journal.uin-alauddin.ac.id/index.php/khizanah-al-hikmah/article/download/55528/22309",
"source": "openalex",
"doi": "https://doi.org/10.24252/v13i1a11",
"abstract": "This study explores the influence of digital archival governance on service efficiency, transparency, and accessibility in Makassar City. The data were gathered through semi-structured interviews, observations, and document analysis involving government officials, policymakers, and IT personnel. The findings indicated that the effectiveness of digital archival governance was shaped by institutional preparedness, supportive legal frameworks, community engagement, and technological infrastructure. Capacity-building programs and professional training were essential to enhance the digital competencies of public servants, while participatory governance mechanisms contributed to greater public trust and access. Additionally, emerging technologies such as blockchain and artificial intelligence presented significant opportunities for optimizing archival systems, though they also introduce challenges related to cybersecurity and ethical management. This study contributes to the discourse on smart governance by highlighting both best practices and obstacles encountered in implementing digital transformation within public administration. It concludes by emphasizing the importance of collaborative governance models, adaptive institutional strategies, and ongoing technological innovation to ensure inclusive and sustainable urban development. Makassar’s experience offers valuable insights for other cities aiming to strengthen digital archival governance as a foundation for improved public service delivery."
},
{
"venue": "CHIIR",
"title": "PRISM: A Framework for Producing Interpretable Political Bias Embeddings with Political-Aware Cross-Encoder",
"authors": [
"Yiqun Sun",
"Qiang Huang",
"Anthony K. H. Tung",
"Jun Yu"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.acl-long.1344.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.acl-long.1344",
"abstract": "Semantic Text Embedding is a fundamental NLP task that encodes textual content into vector representations, where proximity in the embedding space reflects semantic similarity.While existing embedding models excel at capturing general meaning, they often overlook ideological nuances, limiting their effectiveness in tasks that require an understanding of political bias.To address this gap, we introduce PRISM, the first framework designed to Produce inteRpretable polItical biaS eMbeddings.PRISM operates in two key stages: (1) Controversial Topic Bias Indicator Mining, which systematically extracts fine-grained political topics and their corresponding bias indicators from weakly labeled news data, and (2) Cross-Encoder Political Bias Embedding, which assigns structured bias scores to news articles based on their alignment with these indicators.This approach ensures that embeddings are explicitly tied to bias-revealing dimensions, enhancing both interpretability and predictive power.Through extensive experiments on two large-scale datasets, we demonstrate that PRISM outperforms stateof-the-art text embedding models in political bias classification while offering highly interpretable representations that facilitate diversified retrieval and ideological analysis."
},
{
"venue": "CHIIR",
"title": "A Tool for Researching how AI Affects Information Seeking",
"authors": [
"Alamir Novin",
"Georgia Towne"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3698204.3716477",
"source": "openalex",
"doi": "https://doi.org/10.1145/3698204.3716477",
"abstract": "Information Scholars are calling for more standardization of the metrics quantifying the information-seeking processes (ISP).Standardizing measures for online searching is challenging due to search pages including the dynamic nature of generative AI adjacent to personalized search results.AI-chatbots occasionally use queries as prompts to generate text using large-language models (LLM).Search engines also occasionally include Retrieval-Augmented Generation (RAG) results adjacent to personalized search results.This interface hinders the level of control in experiments when two similar queries are treated as prompts towards very different results and text-generations.In addition, experiment findings are less comparable when researchers use different combinations of software for data log collections (e.g., click-throughs and time-on-site), analysis (e.g., NVivo and Qualtrics), and visualizations.These different combinations of platforms can lead to different statistical results or even obnubilate the research metrics.To address these discrepancies, this study introduces a method with three key automations to assist with standardizing ISP experiments on search pages with LLMs and RAG: 1) Controls: A system designed for Randomized Control Trial Experiments and A/B Tests by simulating Google's search AI and algorithms; 2) Data Collection: Automatic participant log data collection; 3) Analysis and Visualization: Presentation of statistically significant differences in both quantitative and qualitative data, with results visualized alongside proper formatting (e.g., APA citations of the p-value).Preliminary feedback from Information Scholars has been promising, with 86% expressing sufficient value in the proposed method to consider using the software for their future projects."
},
{
"venue": "CHIIR",
"title": "Automating University Administration: A Systematic Review of Chatbot Applications in Higher Education",
"authors": [
"Prashant",
"Monika Poriye",
"Pradeep Mittal",
"Naveen Sharma"
],
"year": 2025,
"pdf_url": "https://proceedings.aijr.org/index.php/ap/article/download/89/61",
"source": "openalex",
"doi": "https://doi.org/10.21467/proceedings.7.6.36",
"abstract": "As higher education institutions (HEIs) increasingly adopt artificial intelligence technologies, chatbots have emerged as a scalable solution for improving administrative services. While earlier studies focused primarily on pedagogical aspects, the administrative dimensions remain less explored. This systematic literature review analysed 38 peer-reviewed research articles on the role of chatbots in various university administrative functions, including admissions, enrollment, student services, and financial aid, adhering to the PRISMA 2020 framework. The review introduced a functional taxonomy categorising chatbots based on interaction style, technological framework, target user group, and administrative objectives. It highlighted shared implementation trends, emphasises benefits like reduced workloads and enhanced service responsiveness, and identifies ongoing challenges such as privacy concerns, limited multilingual capabilities, and a mismatch between evaluation methods and expected outcomes."
},
{
"venue": "CHIIR",
"title": "Customer Comment Clustering for Kahf Face Wash at Kahf Official Shop Using K-Means Method",
"authors": [
"Gilang Arbiansyah",
"Faizal Haq"
],
"year": 2025,
"pdf_url": "https://journal.aptika.org/index.php/jistics/article/download/23/23",
"source": "openalex",
"doi": "https://doi.org/10.64878/jistics.v1i3.23",
"abstract": "The advancement of information technology has encouraged people to shop more confidently, including for men's skincare products. Although data indicate that men's interest in skincare remains relatively low, sales of Kahf Face Wash show high figures. In this context, consumer reviews on e-commerce platforms serve as a valuable source of information for understanding customer satisfaction and experience. This study aims to group consumer comments on Kahf Face Wash products from the Kahf Official Shop using the K-Means clustering method. A total of 4,966 consumer comments were collected automatically through web crawling techniques. These comments then underwent several text processing stages, including case folding, cleaning, tokenization, normalization, removal of stop words, and stemming. After the cleaning process, 2,431 comments remained for analysis. The textual data was transformed into numerical representations using the TF-IDF method, and the optimal number of clusters was determined using the Elbow method, which indicated the optimal value at k = 3. The clustering results categorized the comments into three groups: purchase experience (1,506 comments), product effectiveness (474 comments), and delivery and service (451 comments). Visualization was conducted using PCA and bar charts to better illustrate the distribution and proportion of comments in each cluster. Evaluation of the clustering results using inertia and the Davies–Bouldin Index revealed that the model effectively grouped the comments with a reasonably high quality. This study makes a significant contribution by helping companies analyze customer behavior through an unsupervised learning approach. This method enables companies to efficiently extract structured insights from unstructured reviews, which can be utilized to enhance service quality, marketing strategies, and future product development."
},
{
"venue": "CHIIR",
"title": "Generating Effective Health-Related Queries for Promoting Reliable Search Results",
"authors": [
"Xiana Carrera",
"Marcos Fernández-Pichel",
"David E. Losada"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730202",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730202",
"abstract": "Misinformation on the Internet poses significant risks to users seeking health information.This paper addresses the challenge of generating effective health-related queries to promote reliable search results.We propose a method leveraging Large Language Models to generate synthetic narratives that guide the creation of alternative queries.These queries are designed to retrieve more helpful and fewer harmful documents compared to those retrieved by the original user queries.We evaluate the effectiveness of these queries using classic and neural retrieval models across multiple datasets, demonstrating promising improvements in retrieving reputable content."
},
{
"venue": "CHIIR",
"title": "Bancos Vetoriais e Modelos de Embedding: Avaliação comparativa de desempenho na recuperação semântica em Língua Portuguesa",
"authors": [
"Patrick Fernandes Rezende Ribeiro",
"Juliane de Lima Pires",
"Patrick Alves Bastos",
"Roberto Rigo",
"H Reis",
"K. Hübner",
"M Casagrande",
"Bruno de Paula Marafiga",
"D. Simba",
"Denise Fukumi Tsunoda"
],
"year": 2025,
"pdf_url": "https://rsdjournal.org/rsd/article/download/49768/38931",
"source": "openalex",
"doi": "https://doi.org/10.33448/rsd-v14i10.49768",
"abstract": "O crescimento do uso de modelos de linguagem de grande escala intensificou a demanda por bancos de dados vetoriais capazes de lidar com representações semânticas de alta dimensionalidade. Este estudo teve como objetivo avaliar comparativamente diferentes combinações entre bancos de dados vetoriais e modelos de embedding multilíngues, considerando sua aplicabilidade à recuperação semântica em língua portuguesa. A pesquisa caracteriza-se como experimental e aplicada, conduzida em ambiente local, estruturada em quatro etapas: construção da base de dados, definição de critérios de seleção, implementação de um pipeline de experimentação e realização de avaliações de relevância, diversidade e eficiência. Foram analisadas métricas clássicas de recuperação de informação (Recall@k e nDCG), além de métricas de diversidade e equilíbrio (α-nDCG e ILD) e indicadores de eficiência computacional (latência média, latência p95, uso médio de CPU, uso de RAM e Queries per Second - QPS). Os resultados mostraram que soluções como Milvus e Weaviate se destacam em cenários de maior demanda computacional, enquanto pgvector se mostrou mais eficiente em termos de memória. Alternativas como Chroma e pgvector, demonstraram viabilidade em contextos de menor escala. Entre os modelos de embedding, observou-se desempenho consistente dos modelos multilíngues disponíveis no Hugging Face para tarefas em português. Como contribuição, este trabalho apresenta uma análise empírica sistemática que evidencia as potencialidades e limitações de combinações banco vetorial/embedding, oferecendo subsídios para decisões práticas em projetos de curadoria digital, observatórios de dados e sistemas de recomendação em língua portuguesa."
},
{
"venue": "CHIIR",
"title": "Thinking Smarter, not Harder? Google NotebookLM's Misalignment Problem in Education",
"authors": [
"Christa Albrecht-Crane"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3711670.3764628",
"source": "openalex",
"doi": "https://doi.org/10.1145/3711670.3764628",
"abstract": "This paper examines Google's NotebookLM as a case study of how consumer-facing generative AI technologies misalign with educational values and user needs. While marketed as an \"AI-powered research assistant,\" NotebookLM exemplifies the disconnect between AI industry promises and actual capabilities. Through technical analysis of large language model mechanisms, this paper reveals how NotebookLM's statistical compression methods fundamentally differ from human cognitive processes of reading and analysis. The paper argues that despite claims of source-grounding, NotebookLM produces outputs that compress rather than comprehend texts, often missing crucial arguments and generating confabulated content. Drawing on examinations of \"smart\" technology rhetoric and extreme usability design, the analysis demonstrates how the tool's frictionless interface obscures computational limitations while potentially undermining cognitive development. The paper concludes by advocating for critical AI literacy in writing studies and technical communication, proposing pedagogical approaches that demystify AI hype and preserve the essential friction necessary for meaningful learning."
},
{
"venue": "CHIIR",
"title": "Towards Supporting Children's Metacomprehension During Web Search",
"authors": [
"Christine Pinney"
],
"year": 2025,
"pdf_url": "https://scholarworks.boisestate.edu/cgi/viewcontent.cgi?article=3490&context=td",
"source": "openalex",
"doi": "https://doi.org/10.18122/td.2352.boisestate",
"abstract": "As children interact online with search engine result pages (SERPs), children’s comprehension monitoring skills are put to work. Children are known to struggle with utilizing online information as the text content can be misaligned with their reading abilities. Comprehension monitoring skills allow people to assess the comprehensibility of text and is measured by how well comprehension predictions align with performance on comprehension tests. Previous research has shown that augmenting standard SERPs with readability visual cues can be helpful for children searching online. While comprehension predictions are traditionally collected after reading, SERP information provides the opportunity to measure children's perceived comprehensibility of online content before they read. In this work, we explore and analyze the relationship between children's comprehension predictions, SERP information, and SERP navigation. Outcomes reveal that participants tend towards overconfidence in SERP-based predictions, but the utilization of certain SERP information when making predictions results in improved prediction accuracy."
},
{
"venue": "CHIIR",
"title": "2nd Workshop on Information Retrieval for Understudied Users (IR4U2) - Bridging User-centered AI with IR: Making Information Retrieval Accessible for All",
"authors": [
"Noemi Mauro",
"Angelo Geninatti Cossatin",
"Maria Soledad Pera",
"Federica Cena",
"Monica Landoni",
"Theo Huibers",
"Emiliana Murgia"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730356",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730356",
"abstract": "The Workshop on Information Retrieval for Understudied Users (IR4U2) serves as a platform to highlight information retrieval (IR) research that directly impacts often understudied user groups. The second (IR4U2) workshop focuses on a user-centred AI perspective, which is vital for informing the design, development, and assessment of information retrieval systems that thoughtfully address the diverse needs of understudied populations, ensuring genuine accessibility and inclusivity. The objectives of IR4U2 are: (1) to build community and awareness by sharing AI and IR developments that serve underrepresented user groups in this research area; (2) to identify challenges and open issues along with lessons learned and challenges inherent to this area of research; and (3) to spark discussions that establish common frameworks for future research."
},
{
"venue": "CHIIR",
"title": "User Studies in Generative Interactive IR ( GenIIR ): An ISIC ‐Informed Systematic Review",
"authors": [
"Afeng Wang",
"Daqing He",
"Zhimeng Luo",
"Feicheng Ma"
],
"year": 2025,
"pdf_url": "https://onlinelibrary.wiley.com/doi/pdfdirect/10.1002/pra2.1515",
"source": "openalex",
"doi": "https://doi.org/10.1002/pra2.1515",
"abstract": "ABSTRACT This poster presents a systematic review of 98 user studies in Generative Interactive Information Retrieval (GenIIR), exploring how users engage with GenAI in interactive information‐seeking contexts. Guided by the Information Seeking in Context (ISIC) framework, we examined two key dimensions: Research Focus and Information Context . Our results show that most studies examined user perceptions and information behavior, while many also addressed system design evaluation. These studies covered diverse domains and user groups. The most common settings were general, health, and education domains. Participants were primarily general users or students, and these studies often involved public tools like ChatGPT. Other studies focused on professional domains and custom GenAI systems, where interaction context, user roles, and task environments were more specialized, highlighting ISIC's emphasis on context‐sensitive information behavior. These patterns reflect the accessibility of public tools and a rising emphasis on context‐sensitive system design. This review provides insights into developing context‐aware, human‐centered GenAI systems."
},
{
"venue": "CHIIR",
"title": "Marcel: A Lightweight and Open-Source Conversational Agent for University Student Support",
"authors": [
"Jan Trienes",
"Anastasiia Derzhanskaia",
"Roland Schwarzkopf",
"Markus Mühling",
"Jörg Schlötterer",
"Christin Seifert"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.emnlp-demos.13.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.emnlp-demos.13",
"abstract": "Jan Trienes, Anastasiia Derzhanskaia, Roland Schwarzkopf, Markus Mühling, Jörg Schlötterer, Christin Seifert. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: System Demonstrations. 2025."
},
{
"venue": "CHIIR",
"title": "Towards a conversational information seeking process model: Characterizing mixed‐initiative user–agent interaction",
"authors": [
"Shiting Fu",
"Tingting Jiang"
],
"year": 2025,
"pdf_url": "https://onlinelibrary.wiley.com/doi/pdfdirect/10.1002/asi.70035",
"source": "openalex",
"doi": "https://doi.org/10.1002/asi.70035",
"abstract": "Abstract Mixed‐initiative interaction is a key feature of Conversational Information Seeking (CIS), where both users and agents actively participate in dialogue. Given the potential risks associated with agent‐initiative interactions, a CIS process model is needed to guide when and how the agent should take the initiative. Conversation analysis was employed to derive a speech act framework from the 37,268 utterances in the Wizard of Search Engine dataset. Lag sequential analysis was then used to identify adjacency pairs of speech acts with significant transition probabilities. This study identified six user acts, including querying, assessing, elaborating, reformulating, instructing, and deciding, and seven agent acts, including answering, inquiring, checking, eliciting, soliciting, informing, and offering. These acts were connected based on their adjacency, resulting in the CIS process model. This model comprises one fundamental querying‐answering sub‐process and four optional sub‐processes: need negotiation, process collaboration, result evaluation, and query elicitation. The robustness of this model was evaluated on another CIS dataset, ConvSearch, with new behavioral patterns emerging. This study identifies the multi‐stage, iterative, and dynamic nature of mixed‐initiative user–agent interaction in CIS, offers methodological insights into CIS data collection, annotation, and analysis, and provides guidance for the development and evaluation of CIS agents."
},
{
"venue": "CHIIR",
"title": "Information Behavior and Library Resource Use among Generation Z Undergraduates",
"authors": [
"Mad Khir Johari Abdullah Sani",
"Noor Zaidi Sahid",
"Muhamad Azrin Ismail"
],
"year": 2025,
"pdf_url": "https://rsisinternational.org/journals/ijriss/uploads/vol9-iss10-pg5756-5771-202511_pdf.pdf",
"source": "openalex",
"doi": "https://doi.org/10.47772/ijriss.2025.910000472",
"abstract": "This study investigated the information behaviour and use of library resources of Generation Z undergraduate students at Universiti Teknologi MARA (UiTM) Negeri Sembilan. In a world where digital technology is ubiquitous, academic libraries are struggling to remain relevant to Generation Z. The Generation Z students are using online resources and taking responsibility for their own learning. A questionnaire was administered to undergraduate students at UiTM Negeri Sembilan on their information needs, information seeking behaviour, information literacy skills and use of library resources. The results were subjected to quantitative analysis (Cronbach reliability, descriptive statistics, structural equation modelling). The study found that students had high information needs (mean, 3.85/4), high information seeking behaviour (mean, 3.64/5) and high information literacy skills (mean, 3.72/4). The results showed high reliability (Cronbach’s α > 0.78 for all constructs). The results of the structural equation modelling indicated that information literacy skills (β = 0.536, p < 0.001) were the strongest predictors to library use, with information need (β = 0.235, p < 0.001) next, while information-seeking behaviour (β = 0.117, p = 0.021) was the weakest. The study shows that high information needs does not lead to high library use. Actively and strategically seeking information behaviour is the predictor to library use. If library use is to be increased then libraries at UiTM must develop services which are digital, which provide specific training in information literacy and which supply library resources in formats which are consonant with student information behaviour patterns. The results of this study will serve as a base to allow libraries and educators a framework to improve access to and use of academic information via strategies which are user centred."
},
{
"venue": "CHIIR",
"title": "KEIR @ ECIR 2025: The Second Workshop on Knowledge-Enhanced Information Retrieval",
"authors": [
"Zihan Wang",
"Jinyuan Fang",
"Giacomo Frisoni",
"Zhuyun Dai",
"Zaiqiao Meng",
"Gianluca Moro",
"Emine Yılmaz"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2501.11499",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2501.11499",
"abstract": "Pretrained language models (PLMs) like BERT and GPT-4 have become the foundation for modern information retrieval (IR) systems. However, existing PLM-based IR models primarily rely on the knowledge learned during training for prediction, limiting their ability to access and incorporate external, up-to-date, or domain-specific information. Therefore, current information retrieval systems struggle with semantic nuances, context relevance, and domain-specific issues. To address these challenges, we propose the second Knowledge-Enhanced Information Retrieval workshop (KEIR @ ECIR 2025) as a platform to discuss innovative approaches that integrate external knowledge, aiming to enhance the effectiveness of information retrieval in a rapidly evolving technological landscape. The goal of this workshop is to bring together researchers from academia and industry to discuss various aspects of knowledge-enhanced information retrieval."
},
{
"venue": "CHIIR",
"title": "Dark Patterns for Unethical Software Engineering",
"authors": [
"Cesare Pautasso",
"Tatu Pekka Abrahamsson",
"Jyri Anttila",
"Juulia Hakala",
"Anna Ketola",
"Daniel Knappe",
"Vin Lahtinen",
"Timo Liukko",
"Topi-Matti Poranen",
"Manu Ritala",
"Setl",
"David Ifeoluwa Adelani",
"Haotian Mai",
"Fuming Fang",
"H Huy",
"Junichi Nguyen",
"Isao Yamagishi",
"Echizen",
"Ali Al-Kaswan",
"Maliheh Izadi",
"Tyler Amano-Smerling",
"R Fernando",
"Riichiro Andrade",
"Seiji Mizoguchi",
"Isotani",
"Christina Aperjis",
"Ciril Bosch-Rosa",
"Daniel Friedman",
"Bernardo Huberman",
"Markus Appel",
"Fabian Prietzel",
"Leif Azzopardi",
"David Maxwell",
"Martin Halvey",
"Claudia Hauff",
"Leif Azzopardi",
"Guido Zuccon",
"I Shmuel",
"Uri Becher",
"Benoliel",
"Raquel Benbunan-Fich",
"Gerardo Canfora",
"Luigi Cerulo",
"Marta Cimitile",
"Massimiliano Di",
"Penta",
"Chang-Hoan Cho",
"Hongsik John Cheon",
"Reiner Creutzburg",
"Harold Davis",
"Adrienne De",
"Ruiter",
"Michalis Diamantaris",
"Serafeim Moustakas",
"Lichao Sun",
"Sotiris Ioannidis",
"Jason Polakis",
"Virginia Dignum",
"Cory Doctorow",
"D' Christian",
"Kim-Kwang Raymond Orazio",
"Choo",
"Nick Doty",
"Erik Wilde",
"Georgios Doukidis",
"Diomidis Spinellis",
"Christof Ebert",
"Bruno Dyck",
"Rajesh Manchanda",
"Edward W Felten",
"Guo Freeman",
"Karen Wu",
"Nicholas Nower",
"Donghee Yvette Wohn",
"Batya Friedman",
"David Hendry",
"Ftc",
"Bill Gates",
"Tarleton Gillespie",
"Siddharth Daniel G Goldstein",
"Preston Suri",
"Matthew Mcafee",
"Fernando Ekstrand-Abueg",
"Diaz",
"Ben Green",
"Joshua Habgood-Coote",
"T Hansen",
"A Melnikov",
"Masayuki Hatta",
"Sherry He",
"Brett Hollenbeck",
"Davide Proserpio",
"Sherry He",
"Brett Hollenbeck",
"Davide Proserpio",
"Alex Heath",
"D Hilbert",
"D Redmiles",
"Pamela Hinds",
"Teresa Roberts"
],
"year": 2025,
"pdf_url": "https://www.hillside.net/plop/2024/papers/proceedings/papers/20-pautasso.pdf",
"source": "openalex",
"doi": "https://doi.org/10.64346/plop2024p20",
"abstract": "Unethical software engineers write software to satisfy harmful requirements.While patterns promote beneficial solutions to recurring problems, dark patterns intentionally introduce harmful solutions.In this paper we present a collection of 16 dark patterns widely used by unethical software engineers to violate users privacy (email pixel injector, stealthy input logger), pursue monetization at all costs (aggressive advertiser, ad-blocker detector, pay to win, artificial scarcity hoarder, DRM rug puller, obsolescence planner), commit digital frauds (cybersquatter, sneaky terms degrader, interoperability breaker), manipulate search rankings (fake review generator, search ranking kickbacker), and engage in unethical artificial intelligence practices (training data harvester, bot pretender, deceptive deepfaker).By discussing the ethical consequences of each pattern we aim to raise awareness about them and encourage their avoidance by ethical software engineers, architects and practitioners. CCS Concepts Social and professional topics Codes of ethics; Software and its engineering Design patterns."
},
{
"venue": "CHIIR",
"title": "Generating User Personas for Eliciting Requirements Using Online News Data",
"authors": [
"Halim Wildan Awalurahman",
"Indra Kharisma Raharjana",
"Kartono Kartono",
"Shukor Sanim Mohd Fauzi"
],
"year": 2025,
"pdf_url": "https://e-journal.unair.ac.id/JISEBI/article/download/77942/34428",
"source": "openalex",
"doi": "https://doi.org/10.20473/jisebi.11.3.407-419",
"abstract": "Background: In software development, creating user personas remains challenging despite their recognized value. Cost, adaptability, and data scarcity present obstacles in designing these critical personas. A new perspective and process innovation for generating user personas is essential to overcome this hurdle. Objective: This study introduces a method for extracting user persona attributes, including names, occupations, workplaces, and goals. Methods: A framework for extracting user persona information from online news sources is created. Our method employs a comprehensive SpaCy processing pipeline, incorporating NER, SpaCy rule-based matching, and phrase matching. Results: The evaluation results showcase promising performance metrics, with an average recall value of 0.700, precision of 0.402, and F1-score of 0.506. Conclusion: This study demonstrates the feasibility of extracting user persona attributes from online news data. Future research could focus on enhancing the method’s performance, investigating its effectiveness in creating relationships, and ensuring that the generated user personas accurately reflect the news text data. Keywords: Process innovation, Natural Language Processing, Online News, Software Development, User Persona"
},
{
"venue": "CHIIR",
"title": "Understanding Information Seeking Behaviors in Specialized Agricultural Contexts: The Tutur Apple Farming Case",
"authors": [
"Suryaman Sule",
"Kliwon Hidayat",
"Mangku Purnomo",
"Edi Dwi Cahyono"
],
"year": 2025,
"pdf_url": "https://habitat.ub.ac.id/index.php/habitat/article/download/3799/506",
"source": "openalex",
"doi": "https://doi.org/10.21776/ub.habitat.2025.036.3.18",
"abstract": "Apple farming in the Tutur region faces adaptation challenges due to climate change, environmental degradation, and limited access to information. This study highlights the importance of contextual understanding of farmers' information-seeking behaviour by integrating Social Learning (Bandura) and Community of Practice (Wenger) theories. This study examines how farmers seek, interpret, and disseminate information; the social context that shapes the learning process; and the supporting and inhibiting factors of the community information ecosystem. A qualitative approach with a case study design explored apple farmers' learning practices in depth. Data were obtained through interviews, observation, and documentation, and then analysed thematically with triangulation techniques for validity. Results show that farmers actively seek information through digital media, field experiments, and informal discussions. Learning occurs through observation, imitation, and social reinforcement, and is facilitated by a community of practice that encourages a culture of sharing. However, barriers such as closed-mindedness and institutional weaknesses are still found. This study emphasises the importance of participatory and community-based approaches in developing agricultural information systems. The findings are helpful for extension workers and policymakers to design interventions that are adaptive to farmers' social dynamics."
},
{
"venue": "CHIIR",
"title": "Not a Collaborator or a Supervisor, but an Assistant: Striking the Balance Between Efficiency and Ownership in AI-incorporated Qualitative Data Analysis",
"authors": [
"Anoushka Puranik",
"E. S. Chen",
"Roshan L Peiris",
"Kong, Hidy"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2509.18297",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2509.18297",
"abstract": "Qualitative research offers deep insights into human experiences, but its processes, such as coding and thematic analysis, are time-intensive and laborious. Recent advancements in qualitative data analysis (QDA) tools have introduced AI capabilities, allowing researchers to handle large datasets and automate labor-intensive tasks. However, qualitative researchers have expressed concerns about AI's lack of contextual understanding and its potential to overshadow the collaborative and interpretive nature of their work. This study investigates researchers' preferences among three degrees of delegation of AI in QDA (human-only, human-initiated, and AI-initiated coding) and explores factors influencing these preferences. Through interviews with 16 qualitative researchers, we identified efficiency, ownership, and trust as essential factors in determining the desired degree of delegation. Our findings highlight researchers' openness to AI as a supportive tool while emphasizing the importance of human oversight and transparency in automation. Based on the results, we discuss three factors of trust in AI for QDA and potential ways to strengthen collaborative efforts in QDA and decrease bias during analysis."
},
{
"venue": "CHIIR",
"title": "University aspirations and pathways: experiences of prospective first-in-family high school students from migrant backgrounds",
"authors": [
"Charmian M. Pires",
"Laurie A. Chapin"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s13384-025-00901-w.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s13384-025-00901-w",
"abstract": "Abstract Prospective first-in-family (PFiF) students are high school students whose parents have not attended university. As potential first-generation university students, PFiF students are often considered educationally disadvantaged due to their family’s lack of experience with higher education. The university journey begins in high school, and this qualitative study explored the perspectives of 10 PFiF high school students from migrant backgrounds in Melbourne, Australia. Participants identified key barriers to their university transition, including family inexperience, the school’s inability to anticipate their needs, lack of recognition as a distinct cohort and limited guidance around university processes. However, despite these barriers, participants indicated that though inexperienced, families supported their university ambitions and goals and provided them with encouragement and motivation. Students also reported that connections with high school career advisors, combined with their own personal motivation, resilience and perseverance could facilitate a successful transition to university. Findings significantly highlight the need for greater visibility and identification of a PFiF students’ disadvantage, through school records and government policy. Early intervention programs and easy access to vital resources as well as individualised support while at school, were also identified as important, to build aspirations and assist with university transitions."
},
{
"venue": "CHIIR",
"title": "MDSF: Context-Aware Multi-Dimensional Data Storytelling Framework based on Large language Model",
"authors": [
"Chengze Zhang",
"Changshan Li",
"Shiyang Gao"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2501.01014",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2501.01014",
"abstract": "The exponential growth of data and advancements in big data technologies have created a demand for more efficient and automated approaches to data analysis and storytelling. However, automated data analysis systems still face challenges in leveraging large language models (LLMs) for data insight discovery, augmented analysis, and data storytelling. This paper introduces the Multidimensional Data Storytelling Framework (MDSF) based on large language models for automated insight generation and context-aware storytelling. The framework incorporates advanced preprocessing techniques, augmented analysis algorithms, and a unique scoring mechanism to identify and prioritize actionable insights. The use of fine-tuned LLMs enhances contextual understanding and generates narratives with minimal manual intervention. The architecture also includes an agent-based mechanism for real-time storytelling continuation control. Key findings reveal that MDSF outperforms existing methods across various datasets in terms of insight ranking accuracy, descriptive quality, and narrative coherence. The experimental evaluation demonstrates MDSF's ability to automate complex analytical tasks, reduce interpretive biases, and improve user satisfaction. User studies further underscore its practical utility in enhancing content structure, conclusion extraction, and richness of detail."
},
{
"venue": "RecSys",
"title": "AMIKOM-RECSYS: Enhancing Movie Recommender System using Large Language Model (ChatGpt), Deep Learning and Probabilistic Matrix Factorization",
"authors": [
"Muh Hanafi"
],
"year": 2025,
"pdf_url": "https://bright-journal.org/Journal/index.php/JADS/article/download/897/528",
"source": "openalex",
"doi": "https://doi.org/10.47738/jads.v6i4.897",
"abstract": "E-commerce has become one of the most widely used digital applications globally, enabling personalized product discovery and purchasing. To support these services, recommender systems are essential, offering item suggestions based on user preferences. Most recommender systems rely on machine learning algorithms to estimate user-item relevance scores, often utilizing product ratings. However, a persistent challenge in this domain is the issue of data sparsity, where only a small fraction of users provides explicit ratings, leading to reduced accuracy in recommendation results. In this study, we introduce a novel hybrid recommendation algorithm, named AMIKOM-RECSYS, designed to address the sparsity problem and enhance rating prediction. Our model integrates three main components included a Large Language Model (LLM) using ChatGPT, a Transformer-based encoder (BERT), and Probabilistic Matrix Factorization (PMF). The LLM generates descriptive information about movies based on specific prompts, which is then passed to BERT to encode the content into meaningful 2D vector representations. These enriched embeddings are subsequently utilized by the PMF algorithm to predict missing user-item ratings. We evaluate the proposed model on two benchmark datasets, ML-1M and ML-10M using Root Mean Squared Error (RMSE) as the evaluation metric. The AMIKOM-RECSYS model achieved RMSE values of 0.8681 on ML-1M and 0.7791 on ML-10M under a 50:50 data split, outperforming several baseline models including CNN-PMF, LSTM-PMF, and Attention-PMF. These results highlight the effectiveness of integrating LLM and Transformer-based contextual understanding into matrix factorization frameworks. In future work, we plan to extend this framework by incorporating other matrix factorization techniques such as Singular Value Decomposition (SVD) and integrating additional sources of user information, including social media activity, to further improve recommendation performance."
},
{
"venue": "RecSys",
"title": "Toward Holistic Evaluation of Recommender Systems Powered by Generative Models",
"authors": [
"Yashar Deldjoo",
"Nikhil Mehta",
"Maheswaran Sathiamoorthy",
"Shuai Zhang",
"Pablo Castells",
"Julian McAuley"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730354",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730354",
"abstract": "Recommender systems powered by generative models (Gen-RecSys) extend beyond classical item-ranking by producing open-ended content, which simultaneously unlocks richer user experiences and introduces new risks. On one hand, these systems can enhance personalization and appeal through dynamic explanations and multi-turn dialogues. On the other hand, they might venture into unknown territory-hallucinating nonexistent items, amplifying bias, or leaking private information. Traditional accuracy metrics cannot fully capture these challenges, as they fail to measure factual correctness, content safety, or alignment with user intent."
},
{
"venue": "RecSys",
"title": "The Potential of AutoML for Recommender Systems",
"authors": [
"Tobias Vente",
"Lukas Wegmeth",
"Joeran Beel"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3708319.3734173",
"source": "openalex",
"doi": "https://doi.org/10.1145/3708319.3734173",
"abstract": "Automated Machine Learning (AutoML) has significantly advanced Machine Learning (ML) applications, including model compression, machine translation, and computer vision.Recommender Systems (RecSys) can be seen as an application of ML.Yet AutoML has received little attention from the RecSys community, and RecSys has not received notable attention from the AutoML community.Only a few relatively simple Automated Recommender Systems (AutoRec-Sys) libraries exist that adopt AutoML techniques.However, these libraries are based on student projects and do not offer the features and thorough development of AutoML libraries.We set out to determine how AutoML libraries perform in the scenario of an inexperienced user who wants to implement a recommender system.We compared the predictive performance of 60 AutoML, AutoRecSys, ML, and RecSys algorithms from 15 libraries, including a mean predictor baseline, on 14 explicit feedback RecSys datasets.We found that AutoML and AutoRecSys libraries performed best.AutoML libraries performed best for six of the 14 datasets (43%), but the same AutoML library did not always perform best.The single-best library was the AutoRecSys library Auto-Surprise, which performed best on five datasets (36%).On three datasets (21%), AutoML libraries performed poorly, and RecSys libraries with default parameters performed best.Although while obtaining 50% of all placements in the top five per dataset, RecSys algorithms fall behind AutoML on average.ML algorithms generally performed the worst."
},
{
"venue": "RecSys",
"title": "Reindex-Then-Adapt: Improving Large Language Models for Conversational Recommendation",
"authors": [
"Zhankui He",
"Zhouhang Xie",
"Harald Steck",
"Dawen Liang",
"Rahul Jha",
"Nathan Kallus",
"Julian McAuley"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3701551.3703573",
"source": "openalex",
"doi": "https://doi.org/10.1145/3701551.3703573",
"abstract": "Large Language Models (LLMs) are revolutionizing conversational recommender systems (CRS) by effectively indexing item content, understanding complex conversational contexts, and generating relevant item titles. However, the autoregressive nature of LLMs, which outputs item titles as a long sequence of subtokens, hinders the ability to efficiently obtain and control recommendations across the entire item set. This challenge in calculating probabilities over all items limits LLMs' potential, such as (1) limiting control over recommendation popularities and (2) preventing the synergy of marrying LLMs and traditional recommender systems (RecSys)."
},
{
"venue": "RecSys",
"title": "CSRec: Rethinking Sequential Recommendation from A Causal Perspective.",
"authors": [
"Xiaoyu Liu",
"Jiaxin Yuan",
"Yuhang Zhou",
"Jingling Li",
"Fang Huang",
"Wei Ai"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729940",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729940",
"abstract": "The essence of sequential recommender systems (RecSys) lies in understanding how users make decisions.Most existing approaches frame the task as sequential prediction based on users' historical purchase records.Although effective in capturing users' natural preferences, this formulation falls short in accurately modeling actual recommendation scenarios, particularly in accounting for how unsuccessful recommendations influence future purchases.Furthermore, the impact of the RecSys itself on users' decisions has not been appropriately isolated and quantitatively analyzed.To address these challenges, we propose a novel formulation of sequential recommendation, called Causal Sequential Recommendation.Instead of merely predicting the next item in a sequence, CSRec distinguishes between a user's natural preference and their actual purchasing decision.It predicts both aspects within a sequential context and traces how current decisions are formed and causally influenced by various factors.Applying such a causal framework can isolate the impact of recommender systems on user decisions, thereby opening new avenues for evaluation and design.This includes assessing how different strategies influence users' trust in the system and determining the optimal recommender system to maximize advertising benefits.CSRec can be seamlessly integrated into existing next-prediction-based methodologies.Experimental evaluations on both synthetic and real-world datasets demonstrate that the proposed implementation significantly improves upon state-of-the-art baselines.[code can be accessed here]."
},
{
"venue": "RecSys",
"title": "An evaluation review of user similarity metrics in sparse collaborative filtering datasets",
"authors": [
"Kiriakos Sgardelis",
"Dionisis Margaris",
"Dimitris Spiliotopoulos",
"Costas Vassilakis"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s41060-025-00846-4.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s41060-025-00846-4",
"abstract": "Abstract Collaborative filtering (CF) is one of the most prominent recommender system (RecSys) techniques of the recent years. CF generates rating predictions for the items that the user has not evaluated yet, using the evaluations of users with similar likings to the same items. Therefore, in CF the task of finding these users (which can be considered as reliable recommenders) is of high importance, while this task is especially challenging on sparse datasets. To this end, many user similarity metrics have been introduced and used in the literature, such as the Vector (or Cosine) Similarity metric, the Spearman rank correlation, the Pearson Correlation Coefficient (PCC), and others. For a CF RecSys, the use of the most efficient similarity metric is of great importance. This paper assesses the effectiveness of 15 user similarity metrics in sparse CF datasets, by conducting an extensive set of experiments. These experiments include 10 sparse CF datasets with diverse item domains, two neighbour selection approaches, two rating prediction formulas, and three rating prediction accuracy metrics. The evaluation results show that the metrics that achieve the best prediction results are found to be the Spearman rank correlation, followed by the Adjusted Rand Index, the Constrained PCC, and the Chebysev distance. Interestingly, the most widely used similarity metrics in CF research, i.e. the PCC and the Cosine Similarity, are not among the best performing metrics."
},
{
"venue": "RecSys",
"title": "D-RecSys: A Decentralized Recommendation Framework for Web 3.0-Based Content-Sharing Platforms",
"authors": [
"Utsa Roy",
"Ritoja Mukhopadhyay",
"P. Sharma",
"Nirnay Ghosh"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3771093",
"source": "openalex",
"doi": "https://doi.org/10.1145/3771093",
"abstract": "The transition of the web from centralized to decentralized or distributed architectures offers numerous advantages but also introduces significant challenges. One of the key challenges is user profiling to provide personalization, particularly personalized content recommendations. Traditional centralized recommendation systems rely on aggregated user data and central servers, making them incompatible with the principles of decentralization in Web 3.0. To bridge this gap, we propose D-RecSys , a decentralized recommendation framework specifically designed for Web 3.0-based content-sharing dApps. D-RecSys combines federated learning and clustering algorithms to deliver personalized recommendations while preserving user privacy and anonymity. The framework leverages blockchain technology for trustless coordination, enabling the generation of a global model through a modified block structure and mining algorithm. This structure facilitates the aggregation of local models into intermediate block models and subsequently produces the global model. To validate the effectiveness of D-RecSys , we conducted a number of experiments in a simulated Web 3.0 environment. To ensure the generalization capability of the framework, we used three datasets from different domains, i.e., anime recommendation, e-commerce product recommendation, and cellphone recommendation. The results demonstrate that D-RecSys achieves performance levels comparable to centralized recommendation systems while adhering to the core principles of decentralization, user anonymity, and data privacy."
},
{
"venue": "RecSys",
"title": "Blending Sequential Embeddings, Graphs, and Engineered Features: 4th Place Solution in RecSys Challenge 2025",
"authors": [
"Sergei Makeev",
"Alek Andreev",
"Vladimir Baikalov",
"Vladislav Tytskiy",
"А. В. Красильников",
"Kirill Khrylchenko"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3758126.3758131",
"source": "openalex",
"doi": "https://doi.org/10.1145/3758126.3758131",
"abstract": "This paper describes the 4th-place solution by team ambitious for the RecSys Challenge 2025, organized by Synerise and ACM RecSys, which focused on universal behavioral modeling. The challenge objective was to generate user embeddings effective across six diverse downstream tasks. Our solution integrates (1) a sequential encoder to capture the temporal evolution of user interests, (2) a graph neural network to enhance generalization, (3) a deep cross network to model high-order feature interactions, and (4) performance-critical feature engineering."
},
{
"venue": "RecSys",
"title": "Multi-Stage RecSys with Gated-VIC 2T: GFLU-Enhanced Pre-Ranking Meets VICReg",
"authors": [
"Manu Joseph"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3799830.3799870",
"source": "openalex",
"doi": "https://doi.org/10.1145/3799830.3799870",
"abstract": "In any marketplace, we have a hub for third-party sellers to manage their offerings and access strategic insights. However, the abundance of information across mutiple recommendation channels leads to low engagement due to its complexity. To address this, we propose the Multi-Stage Personalized Recommendation System (MS-RecSys), featuring the innovative Gated-VIC 2T Model. This model enhances the Pre-Ranking Stage by integrating Variance, Invariance, and Covariance regularization with Gated Feature Learning Units, surpassing traditional approaches in capturing user-item interactions. Additionally, a Multi-Layered Ensemble system in the Ranking Stage and an optimized inference pipeline significantly improve recommendation relevance and processing efficiency. Our offline evaluations show consistent and considerable improvement in user engagement, ranking precision,and diversity metrics."
},
{
"venue": "RecSys",
"title": "Responsible RecSys by Design: Approximation Algorithms for Calibrated Recommendations with Sponsored Items",
"authors": [
"Jing Yuan",
"Shaojie Tang",
"Shuzhang Cai",
"Yao Wang"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/ICWSM/article/download/35928/38082",
"source": "openalex",
"doi": "https://doi.org/10.1609/icwsm.v19i1.35928",
"abstract": "Calibrated Recommendation Systems (CRS) balance user preferences with constraints like diversity, fairness, and novelty to create inclusive recommendation lists. However, existing research often overlooks the mandatory inclusion of sponsored items, assuming unrestricted product selection. In practice, sponsored items, paid for by advertisers, must be included, which can conflict with CRS goals when advertisers' priorities misalign with system objectives. This paper addresses this gap by formulating CRS with sponsored items as a combinatorial optimization problem. We develop efficient approximation algorithms to generate the most calibrated recommendation lists while meeting sponsorship requirements."
},
{
"venue": "RecSys",
"title": "RecPS: Privacy Risk Scoring for Recommender Systems",
"authors": [
"Jiajie He",
"Yuechun Gu",
"Keke Chen"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3705328.3748052",
"source": "openalex",
"doi": "https://doi.org/10.1145/3705328.3748052",
"abstract": "RecSys '25: Nineteenth ACM Conference on Recommender Systems Prague Czech Republic September 22 - 26, 2025"
},
{
"venue": "RecSys",
"title": "SGCL: Unifying Self-Supervised and Supervised Learning for Graph Recommendation",
"authors": [
"Weizhi Zhang",
"Liangwei Yang",
"Zihe Song",
"Henry Peng Zou",
"Ke Xu",
"Yuanjie Zhu",
"Philip S. Yu"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3705328.3748036",
"source": "openalex",
"doi": "https://doi.org/10.1145/3705328.3748036",
"abstract": "Recommender systems (RecSys) are essential for online platforms, providing personalized suggestions to users within a vast sea of information.Self-supervised graph learning seeks to harness highorder collaborative filtering signals through unsupervised augmentation on the user-item bipartite graph, primarily leveraging a multi-task learning framework that includes both supervised recommendation loss and self-supervised contrastive loss.However, this separate design introduces additional graph convolution processes and creates inconsistencies in gradient directions due to disparate losses, resulting in prolonged training times and sub-optimal performance.In this study, we introduce a unified framework of Supervised Graph Contrastive Learning for recommendation (SGCL) to address these issues.SGCL uniquely combines the training of recommendation and unsupervised contrastive losses into a cohesive supervised contrastive learning loss, aligning both tasks within a single optimization direction for exceptionally fast training.Extensive experiments on three real-world datasets show that SGCL outperforms state-of-the-art methods, achieving superior accuracy and efficiency."
},
{
"venue": "RecSys",
"title": "Encode Me If You Can: Learning Universal User Representations via Event Sequence Autoencoding",
"authors": [
"Anton Klenitskiy",
"Artem Fatkulin",
"Daria Denisova",
"Anton Pembek",
"Alexey Vasilev"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3758126.3758132",
"source": "openalex",
"doi": "https://doi.org/10.1145/3758126.3758132",
"abstract": "Building universal user representations that capture the essential aspects of user behavior is a crucial task for modern machine learning systems. In real-world applications, a user's historical interactions often serve as the foundation for solving a wide range of predictive tasks, such as churn prediction, recommendations, or lifetime value estimation. Using a task-independent user representation that is effective across all such tasks can reduce the need for task-specific feature engineering and model retraining, leading to more scalable and efficient machine learning pipelines. The goal of the RecSys Challenge 2025 by Synerise was to develop such Universal Behavioral Profiles from logs of past user behavior, which included various types of events such as product purchases, page views, and search queries. We propose a method that transforms the entire user interaction history into a single chronological sequence and trains a GRU-based autoencoder to reconstruct this sequence from a fixed-size vector. If the model can accurately reconstruct the sequence, the latent vector is expected to capture the key behavioral patterns. In addition to this core model, we explored several alternative methods for generating user embeddings and combined them by concatenating their output vectors into a unified representation. This ensemble strategy further improved generalization across diverse downstream tasks and helped our team, ai_lab_recsys, achieve second place in the RecSys Challenge 2025."
},
{
"venue": "RecSys",
"title": "Green Recommender Systems: Understanding and Minimizing the Carbon Footprint of AI-Powered Personalization",
"authors": [
"Lukas Wegmeth",
"Tobias Vente",
"Alan Said",
"Joeran Beel"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3768626",
"source": "openalex",
"doi": "https://doi.org/10.1145/3768626",
"abstract": "As global warming soars, the need to assess and reduce the environmental impact of recommender systems is becoming increasingly urgent. Despite this, the recommender systems community hardly understands, addresses, and evaluates the environmental impact of their work. In this study, we examine the environmental impact of recommender systems research by reproducing typical experimental pipelines. Based on our results, we provide guidelines for researchers and practitioners on how to minimize the environmental footprint of their work and implement green recommender systems — recommender systems designed to minimize their energy consumption and carbon footprint. Our analysis covers 79 papers from the 2013 and 2023 ACM RecSys conferences, comparing traditional “good old-fashioned AI” models with modern deep learning models. We designed and reproduced representative experimental pipelines for both years, measuring energy consumption using a hardware energy meter and converting it into CO 2 equivalents. Our results show that papers utilizing deep learning models emit approximately 42 times more CO 2 equivalents than papers using traditional models. On average, a single deep learning-based paper generates 2,909 kilograms of CO 2 equivalents — more than the carbon emissions of a person flying from New York City to Melbourne or the amount of CO 2 sequestered by one tree over 260 years. This work underscores the urgent need for the recommender systems and wider machine learning communities to adopt green AI principles, balancing algorithmic advancements and environmental responsibility to build a sustainable future with AI-powered personalization."
},
{
"venue": "RecSys",
"title": "On explaining recommendations with Large Language Models: a review",
"authors": [
"Alan Said"
],
"year": 2025,
"pdf_url": "https://www.frontiersin.org/journals/big-data/articles/10.3389/fdata.2024.1505284/pdf",
"source": "openalex",
"doi": "https://doi.org/10.3389/fdata.2024.1505284",
"abstract": "The rise of Large Language Models (LLMs), such as LLaMA and ChatGPT, has opened new opportunities for enhancing recommender systems through improved explainability. This paper provides a systematic literature review focused on leveraging LLMs to generate explanations for recommendations-a critical aspect for fostering transparency and user trust. We conducted a comprehensive search within the ACM Guide to Computing Literature, covering publications from the launch of ChatGPT (November 2022) to the present (November 2024). Our search yielded 232 articles, but after applying inclusion criteria, only six were identified as directly addressing the use of LLMs in explaining recommendations. This scarcity highlights that, despite the rise of LLMs, their application in explainable recommender systems is still in an early stage. We analyze these select studies to understand current methodologies, identify challenges, and suggest directions for future research. Our findings underscore the potential of LLMs improving explanations of recommender systems and encourage the development of more transparent and user-centric recommendation explanation solutions."
},
{
"venue": "RecSys",
"title": "Beyond Utility: Evaluating LLM as Recommender",
"authors": [
"Chumeng Jiang",
"Jiayin Wang",
"Weizhi Ma",
"Charles L. A. Clarke",
"Shuai Wang",
"Chuhan Wu",
"Min Zhang"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714759",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714759",
"abstract": "With the rapid development of Large Language Models (LLMs), recent studies employed LLMs as recommenders to provide personalized information services for distinct users. Despite efforts to improve the accuracy of LLM-based recommendation models, relatively little attention is paid to beyond-utility dimensions. Moreover, there are unique evaluation aspects of LLM-based recommendation models, which have been largely ignored. To bridge this gap, we explore four new evaluation dimensions and propose a multidimensional evaluation framework. The new evaluation dimensions include: 1) history length sensitivity, 2) candidate position bias, 3) generation-involved performance, and 4) hallucinations. All four dimensions have the potential to impact performance, but are largely unnecessary for consideration in traditional systems. Using this multidimensional evaluation framework, along with traditional aspects, we evaluate the performance of seven LLM-based recommenders, with three prompting strategies, comparing them with six traditional models on both ranking and re-ranking tasks on four datasets. We find that LLMs excel at handling tasks with prior knowledge and shorter input histories in the ranking setting, and perform better in the re-ranking setting, beating traditional models across multiple dimensions. However, LLMs exhibit substantial candidate position bias issues, and some models hallucinate nonexistent items much more often than others. We intend our evaluation framework and observations to benefit future research on the use of LLMs as recommenders. The code and data are available at https://github.com/JiangDeccc/EvaLLMasRecommender."
},
{
"venue": "RecSys",
"title": "Toward Holistic Evaluation of Recommender Systems Powered by Generative Models",
"authors": [
"Yashar Deldjoo",
"Nikhil Mehta",
"Maheswaran Sathiamoorthy",
"Shuai Zhang",
"Pablo Castells",
"Julian McAuley"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2504.06667",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2504.06667",
"abstract": "Recommender systems powered by generative models (Gen-RecSys) extend beyond classical item ranking by producing open-ended content, which simultaneously unlocks richer user experiences and introduces new risks. On one hand, these systems can enhance personalization and appeal through dynamic explanations and multi-turn dialogues. On the other hand, they might venture into unknown territory-hallucinating nonexistent items, amplifying bias, or leaking private information. Traditional accuracy metrics cannot fully capture these challenges, as they fail to measure factual correctness, content safety, or alignment with user intent. This paper makes two main contributions. First, we categorize the evaluation challenges of Gen-RecSys into two groups: (i) existing concerns that are exacerbated by generative outputs (e.g., bias, privacy) and (ii) entirely new risks (e.g., item hallucinations, contradictory explanations). Second, we propose a holistic evaluation approach that includes scenario-based assessments and multi-metric checks-incorporating relevance, factual grounding, bias detection, and policy compliance. Our goal is to provide a guiding framework so researchers and practitioners can thoroughly assess Gen-RecSys, ensuring effective personalization and responsible deployment."
},
{
"venue": "RecSys",
"title": "Modeling sustainable city trips: integrating $$\\text {CO}_{2}\\text {e}$$ emissions, popularity, and seasonality into tourism recommender systems",
"authors": [
"Ashmi Banerjee",
"Tunar Mahmudov",
"Emil Adler",
"Fitri Nur Aisyah",
"Wolfgang Wörndl"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s40558-024-00303-1.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s40558-024-00303-1",
"abstract": "Abstract Tourism affects not only the tourism industry but also society and stakeholders such as the environment, local businesses, and residents. Tourism recommender systems (TRS) can be pivotal in promoting sustainable tourism by guiding travelers toward destinations with minimal negative impact. Our paper introduces a composite sustainability indicator for a city trip recommender system based on the users’ starting point and month of travel. This indicator integrates CO $$_2$$ 2 e emissions for different transportation modes and analyses destination popularity and seasonal demand. We quantify city popularity based on user reviews, points of interest, and search trends from Tripadvisor and Google Trends data. To calculate a seasonal demand index, we leverage data from TourMIS and Airbnb. We conducted a user study to explore the fundamental trade-offs in travel decision-making and determine the weights for our proposed indicator. Finally, we demonstrate the integration of this indicator into a TRS, illustrating its ability to deliver sustainable city trip recommendations. This work lays the foundation for future research by integrating sustainability measures and contributing to responsible recommendations by TRS."
},
{
"venue": "RecSys",
"title": "SPRec: Self-Play to Debias LLM-based Recommendation",
"authors": [
"Chongming Gao",
"Ruijun Chen",
"Shuai Yuan",
"Kexin Huang",
"Yuanqing Yu",
"Xiangnan He"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714524",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714524",
"abstract": "Large language models (LLMs) have attracted significant attention in recommendation systems.Current work primarily applies supervised fine-tuning (SFT) to adapt the model for recommendation tasks.However, SFT on positive examples only limits the model's ability to align with user preference.To address this, researchers recently introduced Direct Preference Optimization (DPO), which explicitly aligns LLMs with user preferences using offline preference ranking data.However, we found that DPO inherently biases the model towards a few items, exacerbating the filter bubble issue and ultimately degrading user experience.In this paper, we propose SPRec, a novel self-play framework designed to mitigate over-recommendation and improve fairness without requiring additional data or manual intervention.In each self-play iteration, the model undergoes an SFT step followed by a DPO step, treating offline interaction data as positive samples and the predicted outputs from the previous iteration as negative samples.This effectively re-weights the DPO loss function using the model's logits, adaptively suppressing biased items.Extensive experiments on multiple real-world datasets demonstrate SPRec's effectiveness in enhancing recommendation accuracy and fairness.The code is available via https://github.com/RegionCh/SPRec."
},
{
"venue": "RecSys",
"title": "International Workshop on Online and Adaptive Recommender Systems (OARS 2025)",
"authors": [
"Xiquan Cui",
"Derek Zhiyuan Cheng",
"F. LIU",
"Tao Ye",
"Julian McAuley",
"Vachik S. Dave",
"Stephen Guo"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3711896.3737852",
"source": "openalex",
"doi": "https://doi.org/10.1145/3711896.3737852",
"abstract": "Recommender system (RecSys) plays important roles in helping users navigate, discover, and consume massive and highly-dynamic information. Today, many RecSys solutions deployed in the real world rely on categorical user profiles and/or pre-calculated recommendation actions that stay static during a user session. However, recent trends suggest that RecSys need to model user intent in real time and constantly adapt to meet user needs at the moment or change user behavior in situ. There are three primary drivers for this emerging need of online adaptation. First, in order to meet the increasing demand for a better personalized experience, the personalization dimensions and space will grow larger and larger. It would not be feasible to pre-compute recommended actions for all personalization scenarios beyond a certain scale. Second, in many settings the system does not have user prior history to leverage. Estimating user intent in real time is the only way to personalize. As various consumer privacy laws tighten, it is foreseeable that many businesses will reduce their reliance on static user profiles. Therefore, it makes the modeling of user intent in real time an important research topic. Third, a user's intent often changes within a session and between sessions, and user behavior could shift significantly during dramatic events. Therefore, it is important to investigate more on online and adaptive recommender systems (OARS) that can adapt in real time to meet user needs and be robust against distribution shifts. Every year, the organizers survey the most important topics for OARS and propose a new workshop program. In light of the recent advancement of (multi-modal) LLMs in RecSys, in this new edition, we decide to formally add the new topic of (multi-modal) LLM models in OARS. We will invite experts and papers in the field to disseminate new knowledge and foster further advancements."
},
{
"venue": "RecSys",
"title": "A survey on point-of-interest recommendations leveraging heterogeneous data",
"authors": [
"Zehui Wang",
"Wolfram Höpken",
"Dietmar Jannach"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s40558-024-00301-3.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s40558-024-00301-3",
"abstract": "Abstract Tourism is an important application domain for recommender systems. In this domain, recommender systems are for example tasked with providing personalized recommendations for transportation, accommodation, points-of-interest (POIs), etc. Among these tasks, in particular the problem of recommending POIs that are of likely interest to individual tourists has gained growing attention in recent years. Providing POI recommendations to tourists can however be especially challenging due to the variability of the user’s context. With the rapid development of the Web and today’s multitude of online services, vast amounts of data from various sources have become available, and these heterogeneous data represent a huge potential to better address the challenges of POI recommendation problems. In this work, we provide a survey of published research on the problem of POI recommendation between 2021 and 2023. The literature was surveyed to identify the information types, techniques and evaluation methods employed. Based on the analysis, it was observed that the current research tends to focus on a relatively narrow range of information types and there is a significant potential in improving POI recommendation by leveraging heterogeneous data. As the first information-centric survey on POI recommendation research, this study serves as a reference for researchers aiming to develop increasingly accurate, personalized and context-aware POI recommender systems."
},
{
"venue": "RecSys",
"title": "LLM4Rerank: LLM-based Auto-Reranking Framework for Recommendations",
"authors": [
"Jingtong Gao",
"Bo Chen",
"Xiangyu Zhao",
"Weiwen Liu",
"Xiangyang Li",
"Yichao Wang",
"Wanyu Wang",
"Huifeng Guo",
"Ruiming Tang"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714922",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714922",
"abstract": "Reranking is significant for recommender systems due to its pivotal role in refining recommendation results. Numerous reranking models have emerged to meet diverse reranking requirements in practical applications, which not only prioritize accuracy but also consider additional aspects such as diversity and fairness. However, most of the existing models struggle to strike a harmonious balance between these diverse aspects at the model level. Additionally, the scalability and personalization of these models are often limited by their complexity and a lack of attention to the varying importance of different aspects in diverse reranking scenarios. To address these issues, we propose LLM4Rerank, a comprehensive LLM-based reranking framework designed to bridge the gap between various reranking aspects while ensuring scalability and personalized performance. Specifically, we abstract different aspects into distinct nodes and construct a fully connected graph for LLM to automatically consider aspects like accuracy, diversity, fairness, and more, all in a coherent Chain-of-Thought (CoT) process. To further enhance personalization during reranking, we facilitate a customizable input mechanism that allows fine-tuning of LLM's focus on different aspects according to specific reranking needs. Experimental results on three widely used public datasets demonstrate that LLM4Rerank outperforms existing state-of-the-art reranking models across multiple aspects."
},
{
"venue": "RecSys",
"title": "Towards Carbon Footprint-Aware Recommender Systems for Greener Item Recommendation",
"authors": [
"Raoul Kalisvaart",
"Masoud Mansoury",
"Alan Hanjalić",
"Elvin Isufi"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2503.17201",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2503.17201",
"abstract": "The commodity and widespread use of online shopping are having an unprecedented impact on climate, with emission figures from key actors that are easily comparable to those of a large-scale metropolis. Despite online shopping being fueled by recommender systems (RecSys) algorithms, the role and potential of the latter in promoting more sustainable choices is little studied. One of the main reasons for this could be attributed to the lack of a dataset containing carbon footprint emissions for the items. While building such a dataset is a rather challenging task, its presence is pivotal for opening the doors to novel perspectives, evaluations, and methods for RecSys research. In this paper, we target this bottleneck and study the environmental role of RecSys algorithms. First, we mine a dataset that includes carbon footprint emissions for its items. Then, we benchmark conventional RecSys algorithms in terms of accuracy and sustainability as two faces of the same coin. We find that RecSys algorithms optimized for accuracy overlook greenness and that longer recommendation lists are greener but less accurate. Then, we show that a simple reranking approach that accounts for the item's carbon footprint can establish a better trade-off between accuracy and greenness. This reranking approach is modular, ready to use, and can be applied to any RecSys algorithm without the need to alter the underlying mechanisms or retrain models. Our results show that a small sacrifice of accuracy can lead to significant improvements of recommendation greenness across all algorithms and list lengths. Arguably, this accuracy-greenness trade-off could even be seen as an enhancement of user satisfaction, particularly for purpose-driven users who prioritize the environmental impact of their choices. We anticipate this work will serve as the starting point for studying RecSys for more sustainable recommendations."
},
{
"venue": "RecSys",
"title": "Can Users Fix Algorithms? A Game-Theoretic Analysis of Collective Content Amplification in Recommender Systems",
"authors": [
"Ekaterina Fedorova",
"Madeline Kitch",
"Chara Podimata"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2506.04525",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2506.04525",
"abstract": "Users of social media platforms based on recommendation systems (e.g. TikTok, X, YouTube) strategically interact with platform content to influence future recommendations. On some such platforms, users have been documented to form large-scale grassroots movements encouraging others to purposefully interact with algorithmically suppressed content in order to counteractively ``boost'' its recommendation. However, despite widespread documentation of this phenomenon, there is little theoretical work analyzing its impact on the platform or users themselves. We study a game between users and a RecSys, where users (potentially strategically) interact with the content available to them, and the RecSys -- limited by preference learning ability -- provides each user her approximately most-preferred item. We compare recommendations and social welfare when users interact with content according to their personal interests and when a collective of users intentionally interacts with an otherwise suppressed item. We provide sufficient conditions to ensure a pareto improvement in recommendations and strict increases in user social welfare under collective interaction, and provide a robust algorithm to find an effective collective strategy. Interestingly, despite the intended algorithmic protest of these movements, we show that for commonly assumed recommender utility functions, effective collective strategies also improve the utility of the RecSys. Our theoretical analysis is complemented by empirical results of effective collective interaction strategies on the GoodReads dataset and an online survey on how real-world users attempt to influence others' recommendations on RecSys platforms. Our findings examine how and when platforms' recommendation algorithms may incentivize users to collectivize and interact with content in algorithmic protest as well as what this collectivization means for the platform."
},
{
"venue": "RecSys",
"title": "ChatCRS: Incorporating External Knowledge and Goal Guidance for LLM-based Conversational Recommender Systems",
"authors": [
"Chuang Li",
"Yang Deng",
"Hengchang Hu",
"Min‐Yen Kan",
"Haizhou Li"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-naacl.17.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-naacl.17",
"abstract": "We enable large language models (LLMs) to efficiently use external knowledge and goal guidance in conversational recommender system (CRS) tasks.LLMs currently achieve limited effectiveness in domain-specific CRS tasks for 1) generating grounded responses with recommendation-oriented knowledge, or 2) proactively leading the conversations through different dialogue goals.We analyze these limitations through a comprehensive evaluation, showing the necessity of external knowledge and goal guidance which contribute significantly to the recommendation accuracy and language quality.This finding leads us to propose the ChatCRS framework, which decomposes the complex task of CRS into sub-tasks through the implementation of 1) a knowledge retrieval agent using a tool-augmented approach to reason over external knowledge bases, and 2) a goal-planning agent for dialogue goal prediction.By incorporating these inputs, LLMs proactively plan interactions and generate outputs with rich information.Experiments on two multi-goal CRS datasets reveal that ChatCRS sets new state-of-the-art performance, improving language quality of informativeness by 17% and proactivity by 27%, with a tenfold recommendation accuracy enhancement 1 ."
},
{
"venue": "RecSys",
"title": "RecPS: Privacy Risk Scoring for Recommender Systems",
"authors": [
"Jiajie He",
"Yuechun Gu",
"Keke Chen"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3705328.3748052.",
"source": "openalex",
"doi": "https://doi.org/10.13016/m2ml9t-br45",
"abstract": "Recommender systems (RecSys) have become an essential component of many web applications. The core of the system is a recommendation model trained on highly sensitive user-item interaction data. While privacy-enhancing techniques are actively studied in the research community, the real-world model development still depends on minimal privacy protection, e.g., via controlled access. Users of such systems should have the right to choose not to share highly sensitive interactions. However, there is no method allowing the user to know which interactions are more sensitive than others. Thus, quantifying the privacy risk of RecSys training data is a critical step to enabling privacy-aware RecSys model development and deployment. We propose a membership-inference attack (MIA)- based privacy scoring method, RecPS, to measure privacy risks at both the interaction and user levels. The RecPS interaction-level score definition is motivated and derived from differential privacy, which is then extended to the user-level scoring method. A critical component is the interaction-level MIA method RecLiRA, which gives high-quality membership estimation. We have conducted extensive experiments on well-known benchmark datasets and RecSys models to show the unique features and benefits of RecPS scoring in risk assessment and RecSys model unlearning."
},
{
"venue": "RecSys",
"title": "Health recommender systems to facilitate collaborative decision-making in chronic disease management: A scoping review",
"authors": [
"Antonia Barbaric",
"Kenneth Christofferson",
"Susanne M. Benseler",
"Chitra Lalloo",
"Alex Mariakakis",
"Quỳnh Phạm",
"Joost F. Swart",
"Rae S. M. Yeung",
"Joseph A Cafazzo"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1177/20552076241309386",
"source": "openalex",
"doi": "https://doi.org/10.1177/20552076241309386",
"abstract": "Objective: Health recommender systems (HRSs) are increasingly used to complement existing clinical decision-making processes, but their use for chronic diseases remains underexplored. Recognizing the importance of collaborative decision making (CDM) and patient engagement in chronic disease treatment, this review explored how HRSs support patients in managing their illness. Methods: A scoping review was conducted using the framework proposed by Arksey and O'Malley, advanced by Levac et al., in line with the PRISMA-ScR checklist. Quantitative (descriptive numerical summary) and qualitative (inductive content analysis) methods wered used to synthesize the data. Results: Forty-five articles were included in the final review, most commonly covering diabetes (9/45, 20%), mental health (9/45, 20.0%), and tobacco dependence (7/45, 15.6%). Behavior change theories (10/45, 22.2%) and authoritative sources (10/45, 22.2%) were the most commonly referenced sources for design and development work. From the thematic analysis, we conclude: (a) the main goal of HRSs is to induce behavior change, but limited research investigates their effectiveness in achieving this aim; (b) studies acknowledge that theories, models, frameworks, and/or guidelines help design HRSs to elicit specific behavior change, but they do not implement them; (c) connections between CDM and HRS purpose should be more explicit; and (d) HRSs can often offer other self-management services, such as progress tracking and chatbots. Conclusions: We recommend a greater emphasis on evaluation outcomes beyond algorithmic performance to determine HRS effectiveness and the creation of an evidence-driven, methodological approach to creating HRSs to optimize their use in enhancing patient care. Lay summary: Our work aims to provide a summary of the current landscape of health recommender system (HRS) use for chronic disease management. HRSs are digital tools designed to help people manage their health by providing personalized recommendations based on their health history, behaviors, and preferences, enabling them to make more informed health decisions. Given the increased use of these tools for personalized care, and especially with advancements in generative artificial intelligence, understanding the current methods and evaluation processes used is integral to optimizing their effectiveness. Our findings show that HRSs are most used for diabetes, mental health, and tobacco dependence, but only a small percentage of publications directly reference and/or use relevant frameworks to help guide their design and evaluation processes. Furthermore, the goal for most of these HRSs is to induce behavior change, but there is limited research investigating how effective they are in accomplishing this. Given these findings, we recommend that evaluations shift their focus from algorithms to more holistic approaches and to be more intentional about the processes used when designing the tool to support an evidence-driven approach and ultimately create more effective and useful HRSs for chronic disease management."
},
{
"venue": "RecSys",
"title": "Time to Split: Exploring Data Splitting Strategies for Offline Evaluation of Sequential Recommenders",
"authors": [
"Danil Gusak",
"Anna Volodkevich",
"Anton Klenitskiy",
"Alexey Vasilev",
"Evgeny Frolov"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3705328.3748164",
"source": "openalex",
"doi": "https://doi.org/10.1145/3705328.3748164",
"abstract": "Modern sequential recommender systems, ranging from lightweight transformer-based variants to large language models, have become increasingly prominent in academia and industry due to their strong performance in the next-item prediction task.Yet common evaluation protocols for sequential recommendations remain insufficiently developed: they often fail to reflect the corresponding recommendation task accurately, or are not aligned with real-world scenarios.Although the widely used leave-one-out split matches next-item prediction, it permits the overlap between training and test periods, which leads to temporal leakage and unrealistically long test horizon, limiting real-world relevance.Global temporal splitting addresses these issues by evaluating on distinct future periods.However, its applications to sequential recommendations remain"
},
{
"venue": "RecSys",
"title": "TELEClass: Taxonomy Enrichment and LLM-Enhanced Hierarchical Text Classification with Minimal Supervision",
"authors": [
"Yunyi Zhang",
"Ruozhen Yang",
"Xueqiang Xu",
"Rui Li",
"Jinfeng Xiao",
"Jiaming Shen",
"Jiawei Han"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714940",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714940",
"abstract": "Hierarchical text classification aims to categorize each document into a set of classes in a label taxonomy, which is a fundamental web text mining task with broad applications such as web content analysis and semantic indexing. Most earlier works focus on fully or semi-supervised methods that require a large amount of human annotated data which is costly and time-consuming to acquire. To alleviate human efforts, in this paper, we work on hierarchical text classification with a minimal amount of supervision: using the sole class name of each node as the only supervision. Recently, large language models (LLM) have shown competitive performance on various tasks through zero-shot prompting, but this method performs poorly in the hierarchical setting because it is ineffective to include the large and structured label space in a prompt. On the other hand, previous weakly-supervised hierarchical text classification methods only utilize the raw taxonomy skeleton and ignore the rich information hidden in the text corpus that can serve as additional class-indicative features. To tackle the above challenges, we propose TELEClass, Taxonomy Enrichment and LLM-Enhanced weakly-supervised hierarchical text Classification, which combines the general knowledge of LLMs and task-specific features mined from an unlabeled corpus. TELEClass automatically enriches the raw taxonomy with class-indicative features for better label space understanding and utilizes novel LLM-based data annotation and generation methods specifically tailored for the hierarchical setting. Experiments show that TELEClass can significantly outperform previous baselines while achieving comparable performance to zero-shot prompting of LLMs with drastically less inference cost."
},
{
"venue": "RecSys",
"title": "Re2LLM: Reflective Reinforcement Large Language Model for Session-based Recommendation",
"authors": [
"Ziyan Wang",
"Yingpeng Du",
"Zhu Sun",
"Haoyan Chua",
"Kaidong Feng",
"Wenya Wang",
"Jie Zhang"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/33399/35554",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i12.33399",
"abstract": "Emerging advancements in large language models (LLMs) show significant potential for enhancing recommendations. However, prompt-based methods often struggle to find ideal prompts without task-specific feedback, while fine-tuning-based methods are hindered by high computational demands and dependence on open-source backbones. To address these challenges, we propose a Reflective Reinforcement Large Language Model (Re2LLM) for session-based recommendation, which refines LLMs to generate and utilize specialized knowledge effectively and efficiently. Specifically, we first devise the Reflective Exploration Module to extract and present knowledge in a form that LLMs can easily process. This module enables LLMs to reflect on their recommendation mistakes and construct a hint knowledge base to rectify them effectively. Next, we design the Reinforcement Utilization Module to train a lightweight retrieval agent that elicits correct LLM reasoning. This module recognizes hints as signals to facilitate LLM recommendations and learns to select appropriate hints from the constructed knowledge base using task-specific feedback efficiently. Lastly, we conduct experiments on real-world datasets and demonstrate the superiority of our Re2LLM over state-of-the-art methods."
},
{
"venue": "RecSys",
"title": "The Mental World of Large Language Models in Recommendation: A Benchmark on Association, Personalization, and Knowledgeability",
"authors": [
"Guangneng Hu"
],
"year": 2025,
"pdf_url": "https://doi.org/10.48550/arxiv.2512.17389",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2512.17389",
"abstract": "Large language models (LLMs) have shown potential in recommendation systems (RecSys) by using them as either knowledge enhancer or zero-shot ranker. A key challenge lies in the large semantic gap between LLMs and RecSys where the former internalizes language world knowledge while the latter captures personalized world of behaviors. Unfortunately, the research community lacks a comprehensive benchmark that evaluates the LLMs over their limitations and boundaries in RecSys so that we can draw a confident conclusion. To investigate this, we propose a benchmark named LRWorld containing over 38K high-quality samples and 23M tokens carefully compiled and generated from widely used public recommendation datasets. LRWorld categorizes the mental world of LLMs in RecSys as three main scales (association, personalization, and knowledgeability) spanned by ten factors with 31 measures (tasks). Based on LRWorld, comprehensive experiments on dozens of LLMs show that they are still not well capturing the deep neural personalized embeddings but can achieve good results on shallow memory-based item-item similarity. They are also good at perceiving item entity relations, entity hierarchical taxonomies, and item-item association rules when inferring user interests. Furthermore, LLMs show a promising ability in multimodal knowledge reasoning (movie poster and product image) and robustness to noisy profiles. None of them show consistently good performance over the ten factors. Model sizes, position bias, and more are ablated."
},
{
"venue": "RecSys",
"title": "From AutoRecSys to AutoRecLab: A Call to Build, Evaluate, and Govern Autonomous Recommender-Systems Research Labs",
"authors": [
"Joeran Beel",
"Béla Gipp",
"Vente, Tobias",
"Moritz Baumgart",
"Meister, Philipp"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2510.18104",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2510.18104",
"abstract": "Recommender-systems research has accelerated model and evaluation advances, yet largely neglects automating the research process itself. We argue for a shift from narrow AutoRecSys tools -- focused on algorithm selection and hyper-parameter tuning -- to an Autonomous Recommender-Systems Research Lab (AutoRecLab) that integrates end-to-end automation: problem ideation, literature analysis, experimental design and execution, result interpretation, manuscript drafting, and provenance logging. Drawing on recent progress in automated science (e.g., multi-agent AI Scientist and AI Co-Scientist systems), we outline an agenda for the RecSys community: (1) build open AutoRecLab prototypes that combine LLM-driven ideation and reporting with automated experimentation; (2) establish benchmarks and competitions that evaluate agents on producing reproducible RecSys findings with minimal human input; (3) create review venues for transparently AI-generated submissions; (4) define standards for attribution and reproducibility via detailed research logs and metadata; and (5) foster interdisciplinary dialogue on ethics, governance, privacy, and fairness in autonomous research. Advancing this agenda can increase research throughput, surface non-obvious insights, and position RecSys to contribute to emerging Artificial Research Intelligence. We conclude with a call to organise a community retreat to coordinate next steps and co-author guidance for the responsible integration of automated research systems."
},
{
"venue": "RecSys",
"title": "D ata R ec : A Python Library for Standardized and Reproducible Data Management in Recommender Systems",
"authors": [
"Alberto Carlo Maria Mancino",
"Salvatore Bufi",
"Angela Di Fazio",
"Antonio Ferrara",
"Daniele Malitesta",
"Claudio Pomo",
"Tommaso Di Noia"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730320",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730320",
"abstract": "Recommender systems have demonstrated a significant impact across diverse domains, yet ensuring the reproducibility of experimental findings remains a persistent challenge.A primary obstacle lies in the fragmented and often opaque data management strategies employed during the preprocessing stage, where decisions about dataset selection, filtering, and splitting can substantially influence outcomes.To address these limitations, we introduce DataRec, an open-source Python-based library specifically designed to unify and streamline data handling in recommender system research.By providing reproducible routines for dataset preparation, data versioning, and seamless integration with other frameworks, DataRec promotes methodological standardization, interoperability, and comparability across different experimental setups.Our design is informed by an in-depth review of 55 stateof-the-art recommendation studies, ensuring that DataRec adopts best practices while addressing common pitfalls in data management.Ultimately, our contribution facilitates fair benchmarking, enhances reproducibility, and fosters greater trust in experimental results within the broader recommender systems community.The DataRec library, documentation, and examples are freely available at https://github.com/sisinflab/DataRec."
},
{
"venue": "RecSys",
"title": "Unmasking Gender Bias in Recommendation Systems and Enhancing Category-Aware Fairness",
"authors": [
"Tahsin Alamgir Kheya",
"Mohamed Reda Bouadjenek",
"Sunil Aryal"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714528",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714528",
"abstract": "Recommendation systems are now an integral part of our daily lives. We rely on them for tasks such as discovering new movies, finding friends on social media, and connecting job seekers with relevant opportunities. Given their vital role, we must ensure these recommendations are free from societal stereotypes. Therefore, evaluating and addressing such biases in recommendation systems is crucial. Previous work evaluating the fairness of recommended items fails to capture certain nuances as they mainly focus on comparing performance metrics for different sensitive groups. In this paper, we introduce a set of comprehensive metrics for quantifying gender bias in recommendations. Specifically, we show the importance of evaluating fairness on a more granular level, which can be achieved using our metrics to capture gender bias using categories of recommended items like genres for movies. Furthermore, we show that employing a category-aware fairness metric as a regularization term along with the main recommendation loss during training can help effectively minimize bias in the models' output. We experiment on three real-world datasets, using five baseline models alongside two popular fairness-aware models, to show the effectiveness of our metrics in evaluating gender bias. Our metrics help provide an enhanced insight into bias in recommended items compared to previous metrics. Additionally, our results demonstrate how incorporating our regularization term significantly improves the fairness in recommendations for different categories without substantial degradation in overall recommendation performance."
},
{
"venue": "RecSys",
"title": "The Mental World of Large Language Models in Recommendation: A Benchmark on Association, Personalization, and Knowledgeability",
"authors": [
"Guangneng Hu"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2512.17389",
"source": "openalex",
"doi": "",
"abstract": "Large language models (LLMs) have shown potential in recommendation systems (RecSys) by using them as either knowledge enhancer or zero-shot ranker. A key challenge lies in the large semantic gap between LLMs and RecSys where the former internalizes language world knowledge while the latter captures personalized world of behaviors. Unfortunately, the research community lacks a comprehensive benchmark that evaluates the LLMs over their limitations and boundaries in RecSys so that we can draw a confident conclusion. To investigate this, we propose a benchmark named LRWorld containing over 38K high-quality samples and 23M tokens carefully compiled and generated from widely used public recommendation datasets. LRWorld categorizes the mental world of LLMs in RecSys as three main scales (association, personalization, and knowledgeability) spanned by ten factors with 31 measures (tasks). Based on LRWorld, comprehensive experiments on dozens of LLMs show that they are still not well capturing the deep neural personalized embeddings but can achieve good results on shallow memory-based item-item similarity. They are also good at perceiving item entity relations, entity hierarchical taxonomies, and item-item association rules when inferring user interests. Furthermore, LLMs show a promising ability in multimodal knowledge reasoning (movie poster and product image) and robustness to noisy profiles. None of them show consistently good performance over the ten factors. Model sizes, position bias, and more are ablated."
},
{
"venue": "RecSys",
"title": "A Federated Framework for LLM-based Recommendation",
"authors": [
"Jujia Zhao",
"Wenjie Wang",
"C. Xu",
"See-Kiong Ng",
"Tat‐Seng Chua"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-naacl.155.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-naacl.155",
"abstract": "Large Language Models (LLMs) have empowered generative recommendation systems through fine-tuning user behavior data.However, utilizing the user data may pose significant privacy risks, potentially leading to ethical dilemmas and violations of data protection regulations.To address the privacy concerns, Federated Learning for Recommendation (Fed4Rec) has been identified as a promising solution.However, directly applying Fed4Rec in the LLM context introduces two challenges: 1) exacerbated client performance imbalance, which ultimately impacts the system's long-term effectiveness, and 2) substantial client resource costs, posing a high demand for clients' both computational and storage capability to locally train and infer LLMs.To tackle these challenges, we propose a federated framework for LLM-based recommendation (shorted as FELLRec).Generally, FELL-Rec designs two key strategies.1) Dynamic balance strategy, which designs dynamic parameter aggregation and learning speed for different clients, aiming to ensure balanced performance across clients.2) Flexible storage strategy, which selectively retains certain sensitive LLM layers on the client side, while offloading other layers to the server, aiming to preserve privacy while saving resources.Experiment results show that FELLRec can achieve a more balanced client performance and improved overall performance in a computational and storage-efficient way while safeguarding user privacy well."
},
{
"venue": "RecSys",
"title": "Beyond Static LLM Policies: Imitation-Enhanced Reinforcement Learning for Recommendation",
"authors": [
"Yi Zhang",
"Lili Xie",
"Ruihong Qiu",
"Jiajun Liu",
"Sen Wang"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2510.13229",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2510.13229",
"abstract": "Recommender systems (RecSys) have become critical tools for enhancing user engagement by delivering personalized content across diverse digital platforms. Recent advancements in large language models (LLMs) demonstrate significant potential for improving RecSys, primarily due to their exceptional generalization capabilities and sophisticated contextual understanding, which facilitate the generation of flexible and interpretable recommendations. However, the direct deployment of LLMs as primary recommendation policies presents notable challenges, including persistent latency issues stemming from frequent API calls and inherent model limitations such as hallucinations and biases. To address these issues, this paper proposes a novel offline reinforcement learning (RL) framework that leverages imitation learning from LLM-generated trajectories. Specifically, inverse reinforcement learning is employed to extract robust reward models from LLM demonstrations. This approach negates the need for LLM fine-tuning, thereby substantially reducing computational overhead. Simultaneously, the RL policy is guided by the cumulative rewards derived from these demonstrations, effectively transferring the semantic insights captured by the LLM. Comprehensive experiments conducted on two benchmark datasets validate the effectiveness of the proposed method, demonstrating superior performance when compared against state-of-the-art RL-based and in-context learning baselines. The code can be found at https://github.com/ArronDZhang/IL-Rec."
},
{
"venue": "RecSys",
"title": "BEHAV-E! You are Not Just a Number to Us, but an R 2048 Embedding",
"authors": [
"Juan Manuel Rodríguez",
"Antonela Tommasel"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3758126.3758130",
"source": "openalex",
"doi": "https://doi.org/10.1145/3758126.3758130",
"abstract": "Universal user representations hold the promise of reducing the need for task-specific modelling in personalized systems, enabling a single embedding to support a wide range of predictive tasks such as churn prediction, product propensity, and category-level recommendations. In this paper, we introduce BEHAV-E (Behavioural Embedding via Hybrid Action Variational Encoder), a self-supervised architecture that learns rich, 2048-dimensional user embeddings from multi-event behavioural data. BEHAV-E integrates both temporal and semantic signals by combining kernel density estimation (KDE) of user activity timelines, product and category embeddings via shared embedding bags, and an LSTM-based autoencoder for modelling search query semantics. We evaluate our approach within the RecSys Universal Behavioural modelling Challenge. Our results demonstrate that BEHAV-E effectively captures complex, multi-modal user behaviour in a compact and transferable embedding format. This paper provides an overview of the approach we used as team DArgk for the ACM RecSys Challenge 2025. Our submission achieved the 13th rank and sum of scores of 4.5713 (borda count 483) in the competition academia-track final results. We release our source code at: https://github.com/knife982000/RecSys2025Challenge"
},
{
"venue": "RecSys",
"title": "Retrieval-Augmented Purifier for Robust LLM-Empowered Recommendation",
"authors": [
"Liangbo Ning",
"Wenqi Fan",
"Qing Li"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2504.02458",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2504.02458",
"abstract": "Recently, Large Language Model (LLM)-empowered recommender systems have revolutionized personalized recommendation frameworks and attracted extensive attention. Despite the remarkable success, existing LLM-empowered RecSys have been demonstrated to be highly vulnerable to minor perturbations. To mitigate the negative impact of such vulnerabilities, one potential solution is to employ collaborative signals based on item-item co-occurrence to purify the malicious collaborative knowledge from the user's historical interactions inserted by attackers. On the other hand, due to the capabilities to expand insufficient internal knowledge of LLMs, Retrieval-Augmented Generation (RAG) techniques provide unprecedented opportunities to enhance the robustness of LLM-empowered recommender systems by introducing external collaborative knowledge. Therefore, in this paper, we propose a novel framework (RETURN) by retrieving external collaborative signals to purify the poisoned user profiles and enhance the robustness of LLM-empowered RecSys in a plug-and-play manner. Specifically, retrieval-augmented perturbation positioning is proposed to identify potential perturbations within the users' historical sequences by retrieving external knowledge from collaborative item graphs. After that, we further retrieve the collaborative knowledge to cleanse the perturbations by using either deletion or replacement strategies and introduce a robust ensemble recommendation strategy to generate final robust predictions. Extensive experiments on three real-world datasets demonstrate the effectiveness of the proposed RETURN."
},
{
"venue": "RecSys",
"title": "On-Device Recommender Systems: A Comprehensive Survey",
"authors": [
"Hongzhi Yin",
"Zhaojun Li",
"Tong Chen",
"Wei Yuan",
"Ruiqi Zheng",
"Jing Long",
"Xin Xia",
"Yuhui Shi",
"Chengqi Zhang"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s41019-025-00308-8.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s41019-025-00308-8",
"abstract": "Abstract Recommender systems have been widely deployed in various real-world applications to help users identify content of interest from massive amounts of information. Traditional recommender systems work by collecting user-item interaction data in a cloud-based data center and training a centralized model to perform the recommendation service. However, such cloud-based recommender systems (CloudRSs) inevitably suffer from excessive resource consumption, response latency, as well as privacy and security risks concerning both data and models. Recently, driven by the advances in storage, communication, and computation capabilities of edge devices, there has been a shift of focus from CloudRSs to on-device recommender systems (DeviceRSs), which leverage the capabilities of edge devices to minimize centralized data storage requirements, reduce the response latency caused by communication overheads, and enhance user privacy and security by localizing data processing and model training. Despite the rapid rise of DeviceRSs, there is a clear absence of timely literature reviews that systematically introduce, categorize and contrast these methods. To bridge this gap, we aim to provide a comprehensive survey of DeviceRSs, covering three main aspects: (1) the deployment and inference of DeviceRSs, exploring how large recommendation models can be compressed and utilized within resource-constrained on-device environments; (2) the training and update of DeviceRSs, discussing how local data can be leveraged for model optimization on the device side; (3) the security and privacy of DeviceRSs, unveiling their potential vulnerability to malicious attacks and defensive strategies to safeguard these systems. Furthermore, we provide a fine-grained and systematic taxonomy of the methods involved in each aspect, followed by a discussion regarding challenges and future research directions. This is the first comprehensive survey on DeviceRSs that covers a spectrum of tasks to fit various needs. We believe this survey will help readers understand the current research status in this field, equip them with relevant technical foundations, and stimulate new research ideas for developing DeviceRSs."
},
{
"venue": "RecSys",
"title": "Non-parametric Graph Convolution for Re-ranking in Recommendation Systems",
"authors": [
"Zhongyu Ouyang",
"Mingxuan Ju",
"Soroush Vosoughi",
"Yan‐Fang Ye"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3705328.3748058",
"source": "openalex",
"doi": "https://doi.org/10.1145/3705328.3748058",
"abstract": "Graph knowledge has been proven effective in enhancing item rankings in recommender systems (RecSys), particularly during the retrieval stage.However, its application in the ranking stage, especially when richer contextual information in user-item interactions is available, remains underexplored.A major challenge lies in the substantial computational cost associated with repeatedly retrieving neighborhood information from billions of items stored in distributed systems.This resource-intensive requirement makes it difficult to scale graph-based methods in practical RecSys.To bridge this gap, we first demonstrate that incorporating graphs in the ranking stage improves ranking qualities.Notably, while the improvement is evident, we show that the substantial computational overheads entailed by graphs are prohibitively expensive for real-world recommendations.In light of this, we propose a non-parametric strategy that utilizes graph convolution for re-ranking only during test time.Our strategy circumvents the notorious computational overheads from graph convolution during training, and utilizes structural knowledge hidden in graphs on-the-fly during testing.It can be used as a plug-and-play module and easily employed to enhance the ranking ability of various ranking layers of a real-world RecSys with significantly reduced computational overhead.Through comprehensive experiments across four benchmark datasets with varying levels of sparsity, we demonstrate that our strategy yields noticeable improvements (i.e., 8.1% on average) during testing time with little to no additional computational overheads (i.e., 0.5% on average)."
},
{
"venue": "RecSys",
"title": "MLOps Pipelines for Continuous Deployment of Recommendation Systems in Retail",
"authors": [
"Udit Agarwal -",
"Aditya Gupta -"
],
"year": 2025,
"pdf_url": "https://www.ijirct.org/download.php?a_pid=2601004",
"source": "openalex",
"doi": "https://doi.org/10.62970/ijirct.v11.i1.2601004",
"abstract": "The integration of machine learning (ML) models into production environments necessitates a functional operational framework to ensure robustness, scalability, and long-term maintenance. This paper reviews the Machine Learning Operations (MLOps) paradigm as an essential system for the continuous deployment (CD) of personalized recommendation systems (RecSys) within the fast-paced retail sector. MLOps unifies ML development with system operations, facilitating automation across key stages of the model lifecycle from data preparation to deployment and continuous monitoring.2 The framework addresses the unique challenges of retail RecSys, specifically mitigating model drift caused by continuously evolving user preferences and market trends.4 Key architectural components, notably the low-latency Feature Store, are analyzed for their role in maintaining training-serving consistency and enabling real-time inference.6 The paper also examines the critical role of Dynamic Data Management (DDM) integrated into the automated retraining pipeline, which uses data reduction and feature selection to ensure resource-efficient, adaptive model updates. Contemporary approaches to continuous deployment commonly utilize dual metric systems that correlates rank-aware evaluation metrics, such as Normalized Discounted Cumulative Gain (NDCG), with tangible business outcomes like Average Order Value (AOV) lift, alongside operational efficiency metrics such as Mean Time to Resolution (MTTR).9"
},
{
"venue": "RecSys",
"title": "Pantheon: Personalized Multi-objective Ensemble Sort via Iterative Pareto Policy Optimization",
"authors": [
"Jiangxia Cao",
"Pengbo Xu",
"Yin Cheng",
"Kun-feng Guo",
"Jian Tang",
"Shijun Wang",
"Dewei Leng",
"Shuang Yang",
"Zhaojie Liu",
"Yanan Niu",
"Guorui Zhou",
"Kun Gai"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3746252.3761558",
"source": "openalex",
"doi": "https://doi.org/10.1145/3746252.3761558",
"abstract": "To provide promising recommendation results, there exist three major stages in the industrial RecSys chain to support our service: (1) The first Retrieval model aims at searching hundreds of item candidates. (2) Next, the Ranking model estimates the multiple aspect probabilities Pxtrs for each retrieved item. (3) At last, the Ensemble Sort stage merges those Pxtrs into one comparable score, and then selects the best dozen items with the highest scores to recommend them. To our knowledge, the wide-accepted industry ensemble sort approach still relies on manual formula-based adjustment, i.e., assigning manual weights for Pxtrs to control its influence on fusion score. Under this framework, the RecSys severely relies on expert knowledge to determine satisfactory weight for each Pxtr, which blocks RecSys's further advancements."
},
{
"venue": "RecSys",
"title": "Enhancing Recommender Systems: Deep Modality Alignment with Large Multi-Modal Encoders",
"authors": [
"Zixuan Yi",
"Zijun Long",
"Iadh Ounis",
"Craig Macdonald",
"Richard McCreadie"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3718099",
"source": "openalex",
"doi": "https://doi.org/10.1145/3718099",
"abstract": "In recent years, the rapid growth of online multimedia services, such as e-commerce platforms, has necessitated the development of personalised recommendation approaches that can encode diverse content about each item. Indeed, modern multi-modal recommender systems exploit diverse features obtained from raw images and item descriptions to enhance the recommendation performance. However, the existing multi-modal recommender systems primarily depend on the features extracted individually from different media through pre-trained modality-specific encoders, and exhibit only shallow alignments between different modalities, thereby limiting these systems’ ability to capture the underlying relationships between the modalities. In this article, we enhance the deep alignment of large multi-modal encoders to address the shallow alignment of modalities in multi-modal recommender systems. These encoders have previously demonstrated state-of-the-art effectiveness in ranking items across various domains. Specifically, we investigate the use of three state-of-the-art large multi-modal encoders – CLIP (dual-stream), VLMo and BEiT-3 (unified) – for recommendation tasks. We explore their benefits for recommendation through using a range of strategies, including the use of pre-trained and fine-tuned encoders, as well as the evaluation of the end-to-end training of these encoders. We show that pre-trained large multi-modal encoders generate more aligned and effective user/item representations compared with existing modality-specific encoders across four existing multi-modal recommendation datasets. Furthermore, we show that fine-tuning these encoders further improves the recommendation performance, with end-to-end training emerging as the most effective paradigm, significantly outperforming both pre-trained and fine-tuned encoders with an improved recommendation performance. We also demonstrate the effectiveness of large multi-modal encoders in facilitating modality alignment by evaluating the contribution of each modality separately. Finally, we show that the dual-stream approach, specifically CLIP, is the most effective architecture for these large multi-modal encoders, outperforming the unified approaches (i.e., VLMo and BEiT3) in terms of effectiveness and efficiency."
},
{
"venue": "RecSys",
"title": "A novel recommender system using light graph convolutional network and personalized knowledge-aware attention sub-network",
"authors": [
"Rasoul Hassanzadeh",
"Vahid Majidnezhad",
"Bahman Arasteh"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-99949-y.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-99949-y",
"abstract": "Recently, graph neural networks (GNNs) have gained prominence in recommender systems (RS) due to their capability to extract vital features and understand intricate relationships. However, GNNs exhibit limitations in their ability to capture fine-grained semantics in a knowledge graph (KG) and are often insufficient in effectively modeling user-item interactions. One approach to address these limitations is personalized knowledge-aware recommendation. In this paper, a novel RS, called LGKAT, is proposed that uses a combination of user-item graph and knowledge graph. It allows for more precise and nuanced modeling of user-item interactions, aiding the recommender system in learning meaningful node representations. One of the contributions of the proposed method is to use a novel integration of light graph convolutional network (LightGCN) in RSs to efficiently manage common signals for user and item embeddings. Another novelty of this paper is to propose an efficient attention sub-network that encodes rich semantic meanings from the knowledge graph into refined item embeddings in a personalized manner. Extensive tests were conducted on four well-known datasets. The metrics of F1_score and recall is used for the evaluation of the proposed method. The experimental results show the significant superiority of the proposed method compared to the state-of-the-art methods. The obtained results show that the integration of LightGCN in personalized knowledge-aware recommendation systems can effectively tackle limitations of current recommender systems and improve the quality of recommendations."
},
{
"venue": "RecSys",
"title": "Non-parametric Graph Convolution for Re-ranking in Recommendation Systems",
"authors": [
"Zhongyu Ouyang",
"Mingxuan Ju",
"Soroush Vosoughi",
"Yan‐Fang Ye"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2507.09969",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2507.09969",
"abstract": "Graph knowledge has been proven effective in enhancing item rankings in recommender systems (RecSys), particularly during the retrieval stage. However, its application in the ranking stage, especially when richer contextual information in user-item interactions is available, remains underexplored. A major challenge lies in the substantial computational cost associated with repeatedly retrieving neighborhood information from billions of items stored in distributed systems. This resource-intensive requirement makes it difficult to scale graph-based methods in practical RecSys. To bridge this gap, we first demonstrate that incorporating graphs in the ranking stage improves ranking qualities. Notably, while the improvement is evident, we show that the substantial computational overheads entailed by graphs are prohibitively expensive for real-world recommendations. In light of this, we propose a non-parametric strategy that utilizes graph convolution for re-ranking only during test time. Our strategy circumvents the notorious computational overheads from graph convolution during training, and utilizes structural knowledge hidden in graphs on-the-fly during testing. It can be used as a plug-and-play module and easily employed to enhance the ranking ability of various ranking layers of a real-world RecSys with significantly reduced computational overhead. Through comprehensive experiments across four benchmark datasets with varying levels of sparsity, we demonstrate that our strategy yields noticeable improvements (i.e., 8.1% on average) during testing time with little to no additional computational overheads (i.e., 0.5 on average). Code: https://github.com/zyouyang/RecSys2025_NonParamGC.git"
},
{
"venue": "RecSys",
"title": "Order-agnostic Identifier for Large Language Model-based Generative Recommendation",
"authors": [
"Xinyu Lin",
"H. C. Shi",
"Wenjie Wang",
"Fuli Feng",
"Qifan Wang",
"See-Kiong Ng",
"Tat‐Seng Chua"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730053",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730053",
"abstract": "Leveraging Large Language Models (LLMs) for generative recommendation has attracted significant research interest, where item tokenization is a critical step. It involves assigning item identifiers for LLMs to encode user history and generate the next item. Existing approaches leverage either token-sequence identifiers, representing items as discrete token sequences, or single-token identifiers, using ID or semantic embeddings. Token-sequence identifiers face issues such as the local optima problem in beam search and low generation efficiency due to step-by-step generation. In contrast, single-token identifiers fail to capture rich semantics or encode Collaborative Filtering (CF) information, resulting in suboptimal performance."
},
{
"venue": "RecSys",
"title": "Efficient Personalization in E-Commerce: Leveraging Universal Customer Representations with Embeddings",
"authors": [
"Miguel Alves Gomes",
"Philipp Meisen",
"Tobias Meisen"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/0718-1876/20/1/12/pdf?version=1737023886",
"source": "openalex",
"doi": "https://doi.org/10.3390/jtaer20010012",
"abstract": "E-commerce has grown into a billion-dollar industry in recent times with an ever-increasing number of individuals using it regularly. Thus, e-commerce companies can gather interaction data from their customers and analyze it to create focused and personalized marketing campaigns. For large companies, it is possible to tap into these data for personalization using deep learning-based methods that require enormous computing resources. Small companies, on the other hand, cannot afford this. Furthermore, this level of tailor-made addressing necessitates an accurate customer representation. Nevertheless, the exploration of universal representations applicable across various tasks has been limited despite the advantages they offer. We propose a universal customer representation learned only from customer interaction data. We demonstrate that self-supervised trained embeddings of the customer interaction context are a suitable universal customer representation for various e-commerce tasks. To demonstrate the effectiveness of our approach, we conducted experiments comparing four different state-of-the-art approaches and their capabilities in different prediction tasks. Not only do we show that our method outperforms others in most cases, but it also meets other important criteria for real-world applications. It is particularly important to emphasize that our approach does not require a significant amount of resources, and furthermore, is data protection compliant by not using personal information."
},
{
"venue": "RecSys",
"title": "Human-Centered and Sustainable Recommender Systems",
"authors": [
"Allegra De Filippo",
"Ludovico Boratto",
"Giuseppe Spillo"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3708319.3727553",
"source": "openalex",
"doi": "https://doi.org/10.1145/3708319.3727553",
"abstract": "This tutorial explores the intersection of sustainability and recommender systems, focusing on aligning user needs and values with sustainable practices.It emphasizes two dimensions: (1) understanding and modeling users to deliver more sustainable recommendations; and (2) fostering sustainability through system design and functionality.Participants will learn how recommender systems can encourage sustainable behaviors and how to enhance system efficiency while minimizing resource consumption and ethical challenges.Through theoretical insights and hands-on sessions, this tutorial proposes discussion and actionable strategies to design human-centered, sustainable recommender systems, addressing both societal impact and technological responsibility."
},
{
"venue": "RecSys",
"title": "A Survey on LLM-powered Agents for Recommender Systems",
"authors": [
"Qiyao Peng",
"Hongtao Liu",
"Huang Hua",
"Jian Yang",
"Qing Yang",
"Minglai Shao"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.findings-emnlp.620.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.findings-emnlp.620",
"abstract": "Recently, Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language understanding, reasoning, and generation, prompting the recommendation community to leverage these powerful models to address fundamental challenges in traditional recommender systems, including limited comprehension of complex user intents, insufficient interaction capabilities, and inadequate recommendation interpretability.This survey presents a comprehensive synthesis of this rapidly evolving field.We consolidate existing studies into three paradigms: (i) recommenderoriented methods, which directly enhance core recommendation mechanisms; (ii) interactionoriented methods, which conduct multi-turn conversations to elicit preferences and deliver interpretable explanations; and (iii) simulationoriented methods, that model user-item interactions through multi-agent frameworks.Then, we dissect a four-module agent architecture: profile, memory, planning, and action.Then we review representative designs, public datasets, and evaluation protocols.Finally, we give the open challenges that impede real-world deployment, including cost-efficient inference, robust evaluation, and security."
},
{
"venue": "RecSys",
"title": "An Integrated Framework for Contextual Personalized LLM-Based Food Recommendation",
"authors": [
"Ali Rostami"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2504.20092",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2504.20092",
"abstract": "Personalized food recommendation systems (Food-RecSys) critically underperform due to fragmented component understanding and the failure of conventional machine learning with vast, imbalanced food data. While Large Language Models (LLMs) offer promise, current generic Recommendation as Language Processing (RLP) strategies lack the necessary specialization for the food domain's complexity. This thesis tackles these deficiencies by first identifying and analyzing the essential components for effective Food-RecSys. We introduce two key innovations: a multimedia food logging platform for rich contextual data acquisition and the World Food Atlas, enabling unique geolocation-based food analysis previously unavailable. Building on this foundation, we pioneer the Food Recommendation as Language Processing (F-RLP) framework - a novel, integrated approach specifically architected for the food domain. F-RLP leverages LLMs in a tailored manner, overcoming the limitations of generic models and providing a robust infrastructure for effective, contextual, and truly personalized food recommendations."
},
{
"venue": "RecSys",
"title": "Warmer for Less: A Cost-Efficient Strategy for Cold-Start Recommendations at Pinterest",
"authors": [
"Saeed Ebrahimi",
"Weijie Jiang",
"Jaewon Yang",
"Olafur Gudmundsson",
"Yucheng Tu",
"Huizhong Duan"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2512.17277",
"source": "openalex",
"doi": "",
"abstract": "Pinterest is a leading visual discovery platform where recommender systems (RecSys) are key to delivering relevant, engaging, and fresh content to our users. In this paper, we study the problem of improving RecSys model predictions for cold-start (CS) items, which appear infrequently in the training data. Although this problem is well-studied in academia, few studies have addressed its root causes effectively at the scale of a platform like Pinterest. By investigating live traffic data, we identified several challenges of the CS problem and developed a corresponding solution for each: First, industrial-scale RecSys models must operate under tight computational constraints. Since CS items are a minority, any related improvements must be highly cost-efficient. To address this, our solutions were designed to be lightweight, collectively increasing the total parameters by only 5%. Second, CS items are represented only by non-historical (e.g., content or attribute) features, which models often treat as less important. To elevate their significance, we introduce a residual connection for the non-historical features. Third, CS items tend to receive lower prediction scores compared to non-CS items, reducing their likelihood of being surfaced. We mitigate this by incorporating a score regularization term into the model. Fourth, the labels associated with CS items are sparse, making it difficult for the model to learn from them. We apply the manifold mixup technique to address this data sparsity. Implemented together, our methods increased fresh content engagement at Pinterest by 10% without negatively impacting overall engagement and cost, and have been deployed to serve over 570 million users on Pinterest."
},
{
"venue": "RecSys",
"title": "Warmer for Less: A Cost-Efficient Strategy for Cold-Start Recommendations at Pinterest",
"authors": [
"Saeed Ebrahimi",
"Weijie Jiang",
"Jaewon Yang",
"Olafur Gudmundsson",
"Yucheng Tu",
"Huizhong Duan"
],
"year": 2025,
"pdf_url": "https://doi.org/10.48550/arxiv.2512.17277",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2512.17277",
"abstract": "Pinterest is a leading visual discovery platform where recommender systems (RecSys) are key to delivering relevant, engaging, and fresh content to our users. In this paper, we study the problem of improving RecSys model predictions for cold-start (CS) items, which appear infrequently in the training data. Although this problem is well-studied in academia, few studies have addressed its root causes effectively at the scale of a platform like Pinterest. By investigating live traffic data, we identified several challenges of the CS problem and developed a corresponding solution for each: First, industrial-scale RecSys models must operate under tight computational constraints. Since CS items are a minority, any related improvements must be highly cost-efficient. To address this, our solutions were designed to be lightweight, collectively increasing the total parameters by only 5%. Second, CS items are represented only by non-historical (e.g., content or attribute) features, which models often treat as less important. To elevate their significance, we introduce a residual connection for the non-historical features. Third, CS items tend to receive lower prediction scores compared to non-CS items, reducing their likelihood of being surfaced. We mitigate this by incorporating a score regularization term into the model. Fourth, the labels associated with CS items are sparse, making it difficult for the model to learn from them. We apply the manifold mixup technique to address this data sparsity. Implemented together, our methods increased fresh content engagement at Pinterest by 10% without negatively impacting overall engagement and cost, and have been deployed to serve over 570 million users on Pinterest."
},
{
"venue": "RecSys",
"title": "Recommending Paintings in Web Art Gallery with Adjustable Popularity and Diversity",
"authors": [
"Rully Agus Hendrawan",
"Peter Brusilovsky",
"Bereket Abera Yilma",
"Luis A. Leiva"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3708319.3733649",
"source": "openalex",
"doi": "https://doi.org/10.1145/3708319.3733649",
"abstract": "The cold start problem remains a major challenge in visual art recommendation, where limited user feedback often forces systems to rely on content-based filtering.While effective with sufficient data, content similarity-based recommendation can reinforce filter bubbles, narrowing user exposure to mainstream content.Popularity and diversity are both critical factors in recommendation systems, as they impact the visibility of niche items and overall user satisfaction.Yet, existing platforms often rely on popularity-centric algorithms that may discourage exploration and overshadow lesserknown items.To address this gap, our work investigates whether users' preferences for popular and diverse recommendations remain stable over short sessions of recommendation.We propose an interactive, user-adjustable mechanism allowing individuals to control the balance between mainstream and novel suggestions in real-time.We implement this approach within a Web gallery recommendations.Through user study, we examine changes in user behavior.Our findings suggest that while many users initially gravitate toward popular and diverse content, providing controls encourages later adjustments and exploratory behavior.This highlights the need for cultural institutions to move from a tightly managed centralized model to offering users greater affordances for managing the popularity and diversity of personalized recommendations."
},
{
"venue": "RecSys",
"title": "LLMInit: A Free Lunch from Large Language Models for Selective Initialization of Recommendation",
"authors": [
"Weizhi Zhang",
"Liangwei Yang",
"Wooseong Yang",
"Henry Peng Zou",
"Yuqing Liu",
"Ke Xu",
"Sourav Medya",
"Philip S. Yu"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2503.01814",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2503.01814",
"abstract": "Collaborative filtering (CF) is widely adopted in industrial recommender systems (RecSys) for modeling user-item interactions across numerous applications, but often struggles with cold-start and data-sparse scenarios. Recent advancements in pre-trained large language models (LLMs) with rich semantic knowledge, offer promising solutions to these challenges. However, deploying LLMs at scale is hindered by their significant computational demands and latency. In this paper, we propose a novel and scalable LLM-RecSys framework, LLMInit, designed to integrate pretrained LLM embeddings into CF models through selective initialization strategies. Specifically, we identify the embedding collapse issue observed when CF models scale and match the large embedding sizes in LLMs and avoid the problem by introducing efficient sampling methods, including, random, uniform, and variance-based selections. Comprehensive experiments conducted on multiple real-world datasets demonstrate that LLMInit significantly improves recommendation performance while maintaining low computational costs, offering a practical and scalable solution for industrial applications. To facilitate industry adoption and promote future research, we provide open-source access to our implementation at https://github.com/DavidZWZ/LLMInit."
},
{
"venue": "RecSys",
"title": "Learning Universal User Representations Leveraging Cross-domain User Intent at Snapchat",
"authors": [
"Clark Mingxuan Ju",
"Leonardo Neves",
"Bhuvesh Kumar",
"Collins, Liam",
"Tong Zhao",
"Yuwei Qiu",
"Qing Dou",
"Yang Zhou",
"Sohail Nizam",
"Rengim Aykan Ozturk",
"Y. Liu",
"Sen Yang",
"Moona Malik",
"Neil Shah"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2504.21838",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2504.21838",
"abstract": "The development of powerful user representations is a key factor in the success of recommender systems (RecSys). Online platforms employ a range of RecSys techniques to personalize user experience across diverse in-app surfaces. User representations are often learned individually through user's historical interactions within each surface and user representations across different surfaces can be shared post-hoc as auxiliary features or additional retrieval sources. While effective, such schemes cannot directly encode collaborative filtering signals across different surfaces, hindering its capacity to discover complex relationships between user behaviors and preferences across the whole platform. To bridge this gap at Snapchat, we seek to conduct universal user modeling (UUM) across different in-app surfaces, learning general-purpose user representations which encode behaviors across surfaces. Instead of replacing domain-specific representations, UUM representations capture cross-domain trends, enriching existing representations with complementary information. This work discusses our efforts in developing initial UUM versions, practical challenges, technical choices and modeling and research directions with promising offline performance. Following successful A/B testing, UUM representations have been launched in production, powering multiple use cases and demonstrating their value. UUM embedding has been incorporated into (i) Long-form Video embedding-based retrieval, leading to 2.78% increase in Long-form Video Open Rate, (ii) Long-form Video L2 ranking, with 19.2% increase in Long-form Video View Time sum, (iii) Lens L2 ranking, leading to 1.76% increase in Lens play time, and (iv) Notification L2 ranking, with 0.87% increase in Notification Open Rate."
},
{
"venue": "RecSys",
"title": "The Smart Buildings Revolution: A Comprehensive Review of the Smart Readiness Indicator Literature",
"authors": [
"Taraneh Delavar",
"Eerika Borgentorp",
"Seppo Junnila"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2076-3417/15/4/1808/pdf",
"source": "openalex",
"doi": "https://doi.org/10.3390/app15041808",
"abstract": "The construction industry is embracing advanced digital technologies, such as the Internet of Things and automation systems, to enhance energy management and occupant comfort in smart buildings. Recognizing the need to assess the readiness of buildings to support energy-efficient and adaptive functionalities, the European Commission introduced the smart readiness indicator (SRI) in 2018. While the SRI provides a standardized framework, its adoption, limitations, and potential to drive the evolution of smart buildings remain underexplored. This study addresses these gaps through a systematic literature review, incorporating bibliometric and qualitative analyses to evaluate the state of research on the SRI. The bibliometric analysis reveals that research on smart readiness is growing rapidly, with a strong focus on energy efficiency and smart buildings. This literature primarily evaluates and promotes the adoption of the SRI within buildings, aligning with the need to explore the paths for the evolution of smart buildings. The qualitative review summarizes six understudied research topic required to drive the evolution of smart buildings in the literature: The applicability of the SRI to different contexts, including various building types and climatic conditions; the subjectivity in the framework; the alignment with other certificates and standards; the SRI as a tool for smart retrofit; expansion to the neighborhood and district levels; and the score correlation with energy performance. The findings show that, although the SRI was originally introduced for buildings, it has much wider applicability, at the more detailed building component level as well as at the broader neighborhood and district levels. Future research could focus on the role of the SRI in evaluating smart readiness at the neighborhood scale and determining the minimum acceptable SRI score."
},
{
"venue": "RecSys",
"title": "Multi-view knowledge representation learning for personalized news recommendation",
"authors": [
"Chao Chang",
"Feiyi Tang",
"Peng Yang",
"Jingui Zhang",
"Jingxuan Huang",
"Junxian Li",
"Zhenjun Li"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-85166-0.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-85166-0",
"abstract": "In the rapidly evolving field of personalized news recommendation, capturing and effectively utilizing user interests remains a significant challenge due to the vast diversity and dynamic nature of user interactions with news content. Existing recommendation models often fail to fully integrate candidate news items into user interest modeling, which can result in suboptimal recommendation accuracy and relevance. This limitation stems from their insufficient ability to jointly consider user history and the characteristics of candidate news items in the modeling process. To address this challenges, we propose the Multi-view Knowledge Representation Learning (MKRL) framework for personalized news recommendation, which leverages a multi-view news encoder and candidate-aware attention mechanisms to enhance user interest modeling. Unlike traditional methods, MKRL incorporates candidate news articles directly into the user interest modeling process, enabling the model to better understand and predict user preferences based on both historical behavior and potential new content. This is achieved through a sophisticated architecture that blends a multi-view news encoder and candidate-aware attention mechanisms, which together capture a more holistic and dynamic view of user interests. The MKRL framework innovatively integrates convolutional neural networks with multi-head attention modules to capture intricate contextual information from both user history and candidate news, allowing the model to recognize fine-grained patterns. The multi-head attention dynamically weighs user interactions and candidate news based on relevance, enhancing recommendation accuracy. Additionally, MKRL's multi-view approach represents news from different perspectives, enabling richer and more personalized recommendations. Extensive experiments on three real-world datasets demonstrate that our proposed framework outperforms state-of-the-art baselines in recommendation accuracy, validating its effectiveness."
},
{
"venue": "RecSys",
"title": "An Agent‑Based Simulation of Politicized Topics Using Large Language Models: Algorithmic Personalization and Polarization on Social Media",
"authors": [
"Ljubiša Bojić",
"Velibor Ilić",
"Veljko Prodanović",
"Vuk Vuković"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s41111-025-00326-x.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s41111-025-00326-x",
"abstract": "Abstract Digital platforms now act as the primary environments for public discourse, where recommender systems shape visibility, emotion, and interpretation. This study introduces the Recommender Systems LLMs Playground (RecSysLLMsP), a simulation framework designed to examine how algorithmic personalization interacts with language generation to influence engagement and polarization. The research provides a reproducible and transparent environment for testing algorithmic effects on collective reasoning, which is an issue central to democratic communication. The study employs a one‑hundred‑agent simulation grounded in psychometric and demographic data from Serbian social media users. Agents interact through five stages of progressively personalized content feeds mediated by LLM‑generated posts. Quantitative metrics such as engagement intensity, network modularity, sentiment variance and qualitative linguistic validation are used to assess behavioral and structural change. Results reveal that moderate personalization maximizes engagement, while full personalization reduces diversity and amplifies both structural and affective polarization (Q = 0.22 → 0.68). LLM‑based agents successfully reproduce realistic patterns of emotional contagion and ideological clustering. The implications extend to computational social science and policy. Simulation‑based experimentation can inform ethical recommender design and algorithmic governance. Limitations concern the absence of genuine human cognition. Thus, findings indicate systemic tendencies rather than behavioral prediction. Future research should integrate real‑world datasets, multilingual testing, and policy‑driven intervention modeling to further calibrate this digital “laboratory” for exploring AI‑mediated communication."
},
{
"venue": "RecSys",
"title": "Data Access for Recommender Systems Research: leveraging the EU’s Digital Services Act",
"authors": [
"João Vinagre",
"Lorenzo Porcaro",
"Silvia Merisio",
"Erasmo Purificato",
"Emília Gómez"
],
"year": 2025,
"pdf_url": "https://hdl.handle.net/11573/1745386",
"source": "openalex",
"doi": "https://doi.org/10.1145/3705328.3748004",
"abstract": "The European Union (EU) Digital Services Act (DSA) has introduced a novel set of rules for online platforms and search engines, with significant implications for the Recommender Systems community. Through its data access mechanisms, the DSA invites researchers to request both publicly available and private data from Very Large Online Platforms (VLOPs) and Very Large Search Engines (VLOSEs) – those with more than 45 million active recipients in the EU – to investigate systemic risks associated with the dissemination of illegal content, risks to the exercise of fundamental rights, and negative effects on electoral processes, public health, and gender-based violence. This tutorial is aimed at researchers who are interested in submitting such data access requests and will provide them with the knowledge to do so by introducing the relevant definitions and provisions of the DSA, and addressing the most important procedural steps to obtain data access and will provide attendees with a comprehensive understanding of the DSA’s data access implications for RecSys research. The tutorial targets researchers, practitioners, and students in understanding current developments in online platform regulation in Europe and their impact on RecSys research."
},
{
"venue": "RecSys",
"title": "Collaborative Retrieval for Large Language Model-based Conversational Recommender Systems",
"authors": [
"Yaochen Zhu",
"Chao Wan",
"Harald Steck",
"Dawen Liang",
"Yesu Feng",
"Nathan Kallus",
"Jundong Li"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714908",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714908",
"abstract": "Conversational recommender systems (CRS) aim to provide personalized recommendations via interactive dialogues with users. While large language models (LLMs) enhance CRS with their superior understanding of context-aware user preferences, they typically struggle to leverage behavioral data, which have proven to be important for classical collaborative filtering (CF)-based approaches. For this reason, we propose CRAG-Collaborative Retrieval Augmented Generation for LLM-based CRS. To the best of our knowledge, CRAG is the first approach that combines state-of-the-art LLMs with CF for conversational recommendations. Our experiments on two publicly available movie conversational recommendation datasets, i.e., a refined Reddit dataset (which we name Reddit-v2) as well as the Redial dataset, demonstrate the superior item coverage and recommendation performance of CRAG, compared to several CRS baselines. Moreover, we observe that the improvements are mainly due to better recommendation accuracy on recently released movies. The code and data are available at https://github.com/yaochenzhu/CRAG."
},
{
"venue": "RecSys",
"title": "Understanding and Scaling Collaborative Filtering Optimization from the Perspective of Matrix Rank",
"authors": [
"Donald Loveland",
"Xinyi Wu",
"Tong Zhao",
"Danai Koutra",
"Neil Shah",
"Mingxuan Ju"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714904",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714904",
"abstract": "Collaborative Filtering (CF) methods dominate real-world recommender systems given their ability to learn high-quality, sparse ID-embedding tables that effectively capture user preferences. These tables scale linearly with the number of users and items, and are trained to ensure high similarity between embeddings of interacted user-item pairs, while maintaining low similarity for non-interacted pairs. Despite their high performance, encouraging dispersion for non-interacted pairs necessitates expensive regularization (e.g., negative sampling), hurting runtime and scalability. Existing research tends to address these challenges by simplifying the learning process, either by reducing model complexity or sampling data, trading performance for runtime. In this work, we move beyond model-level modifications and study the properties of the embedding tables under different learning strategies. Through theoretical analysis, we find that the singular values of the embedding tables are intrinsically linked to different CF loss functions. These findings are empirically validated on real-world datasets, demonstrating the practical benefits of higher stable rank -- a continuous version of matrix rank which encodes the distribution of singular values. Based on these insights, we propose an efficient warm-start strategy that regularizes the stable rank of the user and item embeddings. We show that stable rank regularization during early training phases can promote higher-quality embeddings, resulting in training speed improvements of up to 65.9%. Additionally, stable rank regularization can act as a proxy for negative sampling, allowing for performance gains of up to 21.2% over loss functions with small negative sampling ratios. Overall, our analysis unifies current CF methods under a new perspective -- their optimization of stable rank -- motivating a flexible regularization method that is easy to implement, yet effective at enhancing CF systems."
},
{
"venue": "RecSys",
"title": "GENNEXT: The Next Generation of IR and Recommender Systems with Language Agents, Generative Models, and Conversational AI",
"authors": [
"Yashar Deldjoo",
"Scott Sanner",
"Enrico Palumbo",
"Hugues Bouchard",
"Shuai Zhang",
"Pablo Castells",
"Julian McAuley"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730369",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730369",
"abstract": "We present GENNEXT, a workshop dedicated to exploring the integration of language agents, generative models, and conversational AI within information retrieval (IR) and recommender systems (RS). Building on the success of our recent RecSys'24 workshop, GENNEXT aims to advance discussions on the applications of language agents powered by Large Language Models (LLMs). The workshop will focus on enhancing interactivity between users and systems through multi-turn dialogues, improving creative content generation, advancing personalization, and enabling multifaceted, context-aware decision-making. For example, a language agent could respond to a query like ''Suggest an eco-friendly food tour for a weekend in my city'' by using a recommendation API to identify eateries specializing in sustainable or organic cuisine and a pollution API to ensure the selected routes have low air pollution levels."
},
{
"venue": "RecSys",
"title": "Enhancing News Recommendation with Hierarchical LLM Prompting",
"authors": [
"Hai-Dang Kieu",
"Delvin Ce Zhang",
"Minh Duc Nguyen",
"Min Xu",
"Qiang Wu",
"Dung D. Le"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3701716.3735085",
"source": "openalex",
"doi": "https://doi.org/10.1145/3701716.3735085",
"abstract": "Personalized news recommendation systems often struggle to effectively capture the complexity of user preferences, as they rely heavily on shallow representations, such as article titles and abstracts. To address this problem, we introduce a novel method, namely PNR-LLM, for Large Language Models for Personalized News Recommendation. Specifically, PNR-LLM harnesses the generation capabilities of LLMs to enrich news titles and abstracts, and consequently improves recommendation quality. PNR-LLM contains a novel module, News Enrichment via LLMs, which generates deeper semantic information and relevant entities from articles, transforming shallow contents into richer representations. We further propose an attention mechanism to aggregate enriched semantic- and entity-level data, forming unified user and news embeddings that reveal a more accurate user-news match. Extensive experiments on MIND datasets show that PNR-LLM outperforms state-of-the-art baselines. Moreover, the proposed data enrichment module is model-agnostic, and we empirically show that applying our proposed module to multiple existing models can further improve their performance, verifying the advantage of our design."
},
{
"venue": "RecSys",
"title": "Intent Representation Learning with Large Language Model for Recommendation",
"authors": [
"Yu Wang",
"Lei Sang",
"Yi Zhang",
"Yiwen Zhang"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730011",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730011",
"abstract": "Intent-based recommender systems have garnered significant attention for uncovering latent fine-grained preferences.Intents, as underlying factors of interactions, are crucial for improving recommendation interpretability.Most methods define intents as learnable parameters updated alongside interactions.However, existing frameworks often overlook textual information (e.g., user reviews, item descriptions), which is crucial for alleviating the sparsity of interaction intents.Exploring these multimodal intents, especially the inherent differences in representation spaces, poses two key challenges: i) How to align multimodal intents and effectively mitigate noise issues; ii) How to extract and match latent key intents across modalities.To tackle these challenges, we propose a modelagnostic framework, Intent Representation Learning with Large Language Model (IRLLRec), which leverages large language models (LLMs) to construct multimodal intents and enhance recommendations.Specifically, IRLLRec employs a dual-tower architecture to learn multimodal intent representations.Next, we propose pairwise and translation alignment to eliminate inter-modal differences and enhance robustness against noisy input features.Finally, to better match textual and interaction-based intents, we employ momentum distillation to perform teacher-student learning on fused intent representations.Empirical evaluations on three datasets show that our IRLLRec framework outperforms baselines 1 ."
},
{
"venue": "RecSys",
"title": "MMAgentRec, a personalized multi-modal recommendation agent with large language model",
"authors": [
"Xiaochen Xiao"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-96458-w.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-96458-w",
"abstract": "In multimodal recommendation, various data types, including text, images, and user dialogues, are utilized. However, it faces two primary challenges. Firstly, identifying user requirements is challenging due to their inherent complexity and diverse intentions. Secondly, the scarcity of high quality datasets and the unnaturalness of recommendation systems pose pressing issues. Especially interactive datasets,and datasets that can evaluate large models and human temporal interactions.In multimodal recommendation, users often face problems such as fragmented information and unclear needs. At the same time, data scarcity affects the accuracy and comprehensiveness of model evaluation and recommendation. This is a pain point in multimodal recommendation. Addressing these issues presents a significant opportunity for advancement. Combining multimodal backgrounds with large language models offers prospects for alleviating pain points. This integration enables systems to support a broader array of inputs, facilitating seamless dialogues and coherent responses. This article employs multimodal techniques, introducing cross-attention mechanisms, self-reflection mechanisms, along with multi-graph neural networks and residual networks. Multimodal techniques are responsible for handling data input problems. Cross-attention mechanisms are used to handle the combination of images and texts. Multi-graph neural networks and residual networks are used to build a recommendation system framework to improve the accuracy of recommendations. These are combined with an adapted large language model (LLM) using the reflection methodology,LLM takes advantage of its ease of communication with humans, proposing an autonomous decision-making and intelligent recommendation-capable multimodal system with self-reflective capabilities. The system includes a recommendation module that seeks advice from different domain experts based on user requirements. Through experimentation, our multimodal system has made significant strides in understanding user intent based on input keywords, demonstrating superiority over classic multimodal recommendation algorithms such as Blip2, clip. This indicates that our system can intelligently generate suggestions, meeting user requirements and enhancing user experience. Our approach provides novel perspectives for the development of multimodal recommendation systems, holding substantial practical application potential and promising to propel their evolution in the information technology domain. This indicates that our system can intelligently generate suggestions, meeting user requirements and enhancing user experience. Our approach provides novel perspectives for the development of multimodal recommendation systems, holding substantial practical application potential and promising to propel their evolution in the information technology domain. We conducted extensive evaluations to assess the effectiveness of our proposed model, including an ablation study, comparison with state-of-the-art methods, and performance analysis on multiple datasets. Ablation Study results demonstrate that the full model achieves the highest performance across all metrics, with an accuracy of 0.9526, precision of 0.94, recall of 0.95, and an F1 score of 0.94. Removing key components leads to performance degradation, with the exclusion of the LLM component having the most significant impact, reducing the F1 score to 0.91. The absence of MGCN and Cross-Attention also results in lower accuracy, confirming their critical role in enhancing model effectiveness. Comparison with state-of-the-art methods indicates that our model outperforms LightGCN and DualGNN in all key metrics. Specifically, LightGCN achieves an accuracy of 0.9210, while DualGNN reaches 0.9285, both falling short of the proposed model's performance. These results validate the superiority of our approach in handling complex multimodal tasks. Experimental results on multiple datasets further highlight the effectiveness of MGCN and Cross-Attention. On the QK-Video and QB-Video datasets, MGCN achieves the highest recall scores, with Recall@5 reaching 0.6556 and 0.6856, and Recall@50 attaining 0.9559 and 0.9059, respectively. Cross-Attention exhibits strong early recall capabilities, achieving Recall@10 of 0.8522 on the Tourism dataset. In contrast, Clip and Blip2 show moderate recall performance, with Clip achieving only 0.3423 for Recall@5 and Blip2 reaching 0.4531 on the Tourism dataset. Overall, our model consistently surpasses existing approaches, with MGCN and Cross-Attention demonstrating superior retrieval and classification performance across various tasks, underscoring their effectiveness in visual question answering (VQA). At the same time, this paper has constructed a comprehensive dataset in this field, each column contains 9004 data entries."
},
{
"venue": "RecSys",
"title": "From Pairwise to Ranking: Climbing the Ladder to Ideal Collaborative Filtering with Pseudo-Ranking",
"authors": [
"Yuhan Zhao",
"Rui Chen",
"Li Chen",
"Shuang Zhang",
"Qilong Han",
"Hongtao Song"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/33462/35617",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i12.33462",
"abstract": "Intuitively, an ideal collaborative filtering (CF) model should learn from users' full rankings over all items to make optimal top-K recommendations. Due to the absence of such full rankings in practice, most CF models rely on pairwise loss functions to approximate full rankings, resulting in an immense performance gap. In this paper, we provide a novel analysis using the multiple ordinal classification concept to reveal the inevitable gap between a pairwise approximation and the ideal case. However, bridging the gap in practice encounters two formidable challenges: (1) none of the real-world datasets contains full ranking information; (2) there does not exist a loss function that is capable of consuming ranking information. To overcome these challenges, we propose a pseudo-ranking paradigm (PRP) that addresses the lack of ranking information by introducing pseudo-rankings supervised by an original noise injection mechanism. Additionally, we put forward a new ranking loss function designed to handle ranking information effectively. To ensure our method's robustness against potential inaccuracies in pseudo-rankings, we equip the ranking loss function with a gradient-based confidence mechanism to detect and mitigate abnormal gradients. Extensive experiments on four real-world datasets demonstrate that PRP significantly outperforms state-of-the-art methods."
},
{
"venue": "RecSys",
"title": "Understanding teams and productivity in information retrieval research: Academia, industry, and cross-community collaborations",
"authors": [
"Jiaqi Lei",
"Liang Hu",
"Yi Bu",
"Jiqun Liu"
],
"year": 2025,
"pdf_url": "https://reference-global.com/2/v2/download/article/10.2478/jdis-2025-0051.pdf",
"source": "openalex",
"doi": "https://doi.org/10.2478/jdis-2025-0051",
"abstract": "ABSTRACT Purpose Prior Information Retrieval (IR) research synthesizes progress from individual studies, yet academia-industry collaboration dynamics remain unexplored. This study investigates: (1) productivity patterns and venues, (2) citations-downloads relationships, (3) topic evolution, and (4) collaboration trends. Design/methodology/approach We perform an analysis of 53,471 ACM IR papers (2000–2018) using bibliometrics and DistilBERT topic modeling. Findings We find that industry-involved papers preferred WWW/CIKM venues; collaborations dominated RecSys/CSCW. We see that academia-industry collaborations achieved the highest download-to-citation conversion rates. Academia focused on algorithms; industry on applications; collaborations bridged both with rising human-centered themes. Research implications This is a pioneering large-scale bibliometrics revealing collaboration’s impact on IR knowledge evolution and provides a methodological framework for cross-sector analysis. Practical implications The paper identifies optimal venues (RecSys/CSCW) for partnerships and guides joint initiatives (shared datasets, grants) to bridge academia-industry divides and enhance research translation. Originality/value This is the first large-scale bibliometric analysis of IR academia-industry collaboration. The paper finds many novel insights, including the fact that collaboration boosts citation efficiency, enables complementary specialization, and drives topic convergence."
},
{
"venue": "RecSys",
"title": "From Sequences to Profiles: Generating Universal Behavioral Profiles exploiting Recurrent Neural Networks",
"authors": [
"Simone Colecchia",
"Mauro Orazio Drago",
"Jihad Founoun",
"Paolo Gennaro",
"Ernesto Natuzzi",
"Luca Pagano",
"Sajjad Shaffaf",
"Giuseppe Vitello",
"Andrea Pisani",
"Maurizio Ferrari Dacrema"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1145/3758126.3758133",
"source": "openalex",
"doi": "https://doi.org/10.1145/3758126.3758133",
"abstract": "This paper presents the solution developed by the EmbedNBreakfast team for the ACM RecSys Challenge 2025, for the construction of Universal Behavioral Profiles: general-purpose user representations derived from historical interactions. We propose a representation-learning framework that combines Recurrent Neural Networks, attention mechanisms, and collaborative filtering to jointly optimize embeddings across several predictive objectives. Our method achieved 2nd place on the Academic Leaderboard and 5th Overall, demonstrating the effectiveness of unified, representation-based modeling for diverse behavior prediction tasks."
},
{
"venue": "RecSys",
"title": "Explainable person–job recommendations: challenges, approaches, and comparative analysis",
"authors": [
"Fang Tang",
"Renqi Zhu",
"Feng Yao",
"Junzhi Wang",
"Lailong Luo",
"Bo Li"
],
"year": 2025,
"pdf_url": "https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2025.1660548/pdf",
"source": "openalex",
"doi": "https://doi.org/10.3389/frai.2025.1660548",
"abstract": "Introduction: As person-job recommendation systems (PJRS) increasingly mediate hiring decisions, concerns over their \"black box\" opacity have sparked demand for explainable AI (XAI) solutions. Methods: This systematic review examines 85 studies on explainable PJRS methods published between 2019 and August 2025, selected from 150 screened articles across Google Scholar, Web of Science, and CNKI, following PRISMA 2020 guidelines. Results: Guided by a PICOS-formulated review question, we categorize explainability techniques into three layers-data (e.g., feature attribution, causal diagrams), model (e.g., attention mechanisms, knowledge graphs), and output (e.g., SHAP, counterfactuals)-and summarize their objectives, trade-offs, and practical applications. We further synthesize these into an integrated end-to-end framework that addresses opacity across layers and supports traceable recommendations. Quantitative benchmarking of six representative methods (e.g., LIME, attention-based, KG-GNN) reveals performance-explainability trade-offs, with counterfactual approaches achieving the highest Explainability-Performance (E‑P) score (0.95). Discussion: This review provides a taxonomy, cross-layer framework, and comparative evidence to inform the design of transparent and trustworthy PJRS systems. Future directions include multimodal causal inference, feedback-driven adaptation, and efficient explainability tools."
},
{
"venue": "RecSys",
"title": "LLMInit: A Free Lunch from Large Language Models for Selective Initialization of Recommendation",
"authors": [
"Association for Computational Linguistics 2025",
"Liu, Yuqing",
"Medya, Sourav",
"S. Yu, Philip",
"Yang, Liangwei",
"Yang, Wooseong",
"Zhang, Weizhi",
"Zou, Henry Peng"
],
"year": 2025,
"pdf_url": "https://doi.org/10.48448/3kn2-hz23",
"source": "openalex",
"doi": "https://doi.org/10.48448/3kn2-hz23",
"abstract": "Collaborative filtering (CF) is widely adopted in industrial recommender systems (RecSys) for modeling user-item interactions across numerous applications, but often struggles with cold-start and data-sparse scenarios. Recent advancements in pre-trained large language models (LLMs) with rich semantic knowledge, offer promising solutions to these challenges. However, deploying LLMs at scale is hindered by their significant computational demands and latency. In this paper, we propose a novel and scalable LLM-RecSys framework, LLMInit, designed to integrate pretrained LLM embeddings into CF models through selective initialization strategies. Specifically, we identify the embedding collapse issue observed when CF models scale and match the large embedding sizes in LLMs and avoid the problem by introducing efficient sampling methods, including, random, uniform, and variance-based selections. Comprehensive experiments conducted on multiple real-world datasets demonstrate that LLMInit significantly improves recommendation performance while maintaining low computational costs, offering a practical and scalable solution for industrial applications. To facilitate industry adoption and promote future research, we provide open-source access to our implementation at https://github.com/DavidZWZ/LLMInit."
},
{
"venue": "RecSys",
"title": "TEARS: Text Representations for Scrutable Recommendations",
"authors": [
"Emiliano Penaloza",
"Olivier Gouvert",
"Haolun Wu",
"Laurent Charlin"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714948",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714948",
"abstract": "Traditional recommender systems rely on high-dimensional (latent) embeddings for modeling user-item interactions, often resulting in opaque representations that lack interpretability.Moreover, these systems offer limited control to users over their recommendations.Inspired by recent work, we introduce TExtuAl Representations for Scrutable recommendations (TEARS) to address these challenges.Instead of representing a user's interests through a latent embedding, TEARS encodes them in natural text, providing transparency and allowing users to edit them.To do so, TEARS uses a modern LLM to generate user summaries based on user preferences.Using these summaries, we take a hybrid approach where we use an optimal transport procedure to align the summaries' representation with the learned representation of a standard VAE for collaborative filtering.We find this approach can surpass the performance of popular VAE models while providing user-controllable recommendations.We also analyze the controllability of TEARS through three simulated user tasks to evaluate the effectiveness of a user editing its summary.A more detailed version of this manuscript with more experiments, baselines and detail is provided on arXiv."
},
{
"venue": "RecSys",
"title": "Collaborative Filtering Algorithm Based on Deep Denoising Auto-Encoder and Attention Mechanism",
"authors": [
"Zixuan Han",
"Leilei Shi",
"Qiang Sun",
"Xiuliang Huang",
"Bing Lei",
"Lu Liu",
"Yao Lu"
],
"year": 2025,
"pdf_url": "https://www.cai.sk/ojs/index.php/cai/article/download/2025_1_176/1340",
"source": "openalex",
"doi": "https://doi.org/10.31577/cai_2025_1_176",
"abstract": "The burgeoning of e-commerce and online platforms has led to an explosion in data volume and diversity of user preferences, making effective recommendation systems crucial for personalizing user experiences. While collaborative filtering algorithms are traditionally favoured for their ability to leverage user-item interactions, they grapple with data sparsity and noise challenges. To tackle these challenges, Various approaches have emerged in recent years to tackle these challenges. Recent strides in deep learning, particularly autoencoders and neural networks, have shown promise in addressing these issues. However, limitations persist, such as suboptimal feature extraction and the underutilization of combined nonlinear and linear latent features in traditional autoencoders, as well as the overlooked impact of active users in recommendations. Addressing these research gaps, this study introduces a novel recommendation algorithm that synergizes a deep denoising autoencoder with an attention mechanism, aiming to refine recommendation performance by mitigating data sparsity and enhancing feature extraction. This fusion approach innovatively combines nonlinear and linear latent features and incorporates a neural attention mechanism, significantly improving the precision and personalization of recommendations. Ultimately, the proposed algorithm's effectiveness is assessed and benchmarked against state-of-the-art approaches, demonstrating its potential to revolutionize recommendation systems by offering more accurate and user-tailored suggestions."
},
{
"venue": "RecSys",
"title": "DARLR: Dual-Agent Offline Reinforcement Learning for Recommender Systems with Dynamic Reward",
"authors": [
"Yi Zhang",
"Ruihong Qiu",
"Xuwei Xu",
"Jiajun Liu",
"Sen Wang"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729942",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729942",
"abstract": "Model-based offline reinforcement learning (RL) has emerged as a promising approach for recommender systems, enabling effective policy learning by interacting with frozen world models. However, the reward functions in these world models, trained on sparse offline logs, often suffer from inaccuracies. Specifically, existing methods face two major limitations in addressing this challenge: (1) deterministic use of reward functions as static look-up tables, which propagates inaccuracies during policy learning, and (2) static uncertainty designs that fail to effectively capture decision risks and mitigate the impact of these inaccuracies. In this work, a dual-agent framework, DARLR, is proposed to dynamically update world models to enhance recommendation policies. To achieve this, a selector is introduced to identify reference users by balancing similarity and diversity so that the recommender can aggregate information from these users and iteratively refine reward estimations for dynamic reward shaping. Further, the statistical features of the selected users guide the dynamic adaptation of an uncertainty penalty to better align with evolving recommendation requirements. Extensive experiments on four benchmark datasets demonstrate the superior performance of DARLR, validating its effectiveness. The code is available at this address."
},
{
"venue": "RecSys",
"title": "AlphaFuse: Learn ID Embeddings for Sequential Recommendation in Null Space of Language Embeddings",
"authors": [
"Guoqing Hu",
"An Zhang",
"Shuo Liu",
"Zhenggang Cai",
"Xun Yang",
"Xiang Wang"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729894",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729894",
"abstract": "Recent advancements in sequential recommendation have underscored the potential of Large Language Models (LLMs) for enhancing item embeddings. However, existing approaches face three key limitations: 1) the degradation of the semantic space when high-dimensional language embeddings are mapped to lower-dimensional ID embeddings, 2) the underutilization of language embeddings, and 3) the reliance on additional trainable parameters, such as an adapter, to bridge the gap between the semantic and behavior spaces. In this paper, we introduce AlphaFuse, a simple but effective language-guided learning strategy that addresses these challenges by learning ID embeddings within the null space of language embeddings. Specifically, we decompose the semantic space of language embeddings via Singular Value Decomposition (SVD), distinguishing it into a semantic-rich row space and a semantic-sparse null space. Collaborative signals are then injected into the null space, while preserving the rich semantics of the row space. AlphaFuse prevents degradation of the semantic space, integrates the retained language embeddings into the final item embeddings, and eliminates the need for auxiliary trainable modules, enabling seamless adaptation to any sequential recommendation framework. We validate the effectiveness and flexibility of AlphaFuse through extensive experiments on three benchmark datasets, including cold-start user and long-tail settings, showcasing significant improvements in both discriminative and diffusion-based generative sequential recommenders."
},
{
"venue": "RecSys",
"title": "Bridging the User-side Knowledge Gap in Knowledge-aware Recommendations with Large Language Models",
"authors": [
"Zheng Hu",
"Zhe Li",
"Ziyun Jiao",
"Satoshi Nakagawa",
"Jiawen Deng",
"Shi‐Min Cai",
"Tao Tang",
"Fuji Ren"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/33284/35439",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i11.33284",
"abstract": "In recent years, knowledge graphs have been integrated into recommender systems as item-side auxiliary information, enhancing recommendation accuracy. However, constructing and integrating structural user-side knowledge remains a significant challenge due to the improper granularity and inherent scarcity of user-side features. Recent advancements in Large Language Models (LLMs) offer the potential to bridge this gap by leveraging their human behavior understanding and extensive real-world knowledge. Nevertheless, integrating LLM-generated information into recommender systems presents challenges, including the risk of noisy information and the need for additional knowledge transfer. In this paper, we propose an LLM-based user-side knowledge inference method alongside a carefully designed recommendation framework to address these challenges. Our approach employs LLMs to infer user interests based on historical behaviors, integrating this user-side information with item-side and collaborative data to construct a hybrid structure: the Collaborative Interest Knowledge Graph (CIKG). Furthermore, we propose a CIKG-based recommendation framework that includes a user interest reconstruction module and a cross-domain contrastive learning module to mitigate potential noise and facilitate knowledge transfer. We conduct extensive experiments on three real-world datasets to validate the effectiveness of our method. Our approach achieves state-of-the-art performance compared to competitive baselines, particularly for users with sparse interactions."
},
{
"venue": "RecSys",
"title": "Model-Agnostic Social Network Refinement with Diffusion Models for Robust Social Recommendation",
"authors": [
"Youchen Sun",
"Zhu Sun",
"Yingpeng Du",
"Jie Zhang",
"Yew-Soon Ong"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714683",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714683",
"abstract": "Social recommendations (SRs) aim to enhance preference modeling by integrating social networks. However, their effectiveness is mainly constrained by two factors: the noisy social connections that may not reflect shared interests, and the limited number of social connections for most users, which hampers the system's ability to fully leverage social influence. Therefore, it is essential to perform social network refinement by removing noisy connections and adding meaningful ones for robust SRs. Inspired by the denoising capability of generative diffusion models, we propose a Model-Agnostic Social Network Refinement framework with Diffusion Models for Robust Social Recommendation (ARD-SR). Specifically, in the forward process, we corrupt the social network by progressively adding position-specific Gaussian noise calibrated to the user preference similarity, better simulating how the social network responds to noise perturbations. The reverse process learns to denoise, guided by each user's neighborhood preferences from the SR backbone, generating a tailored social network aligned with each user's preference for establishing connections. For effective learning, we design a curriculum-based training mechanism that progressively introduces challenging samples characterized by high sparsity or high noise levels. Finally, ARD-SR and the SR backbone are alternately trained, ensuring a continuous mutual enhancement between the social network refinement and the backbone's user representation learning. To further enhance the quality of the refined social network, (1) we introduce a preference-guided flip operation during inference to improve the input quality; and (2) we modify social connections based on the exponential weighted moving average of ARD-SR's predictions across epochs to reduce fluctuations. Experiments on three datasets show that ARD-SR significantly improves SR performance across multiple SR backbones. The code is released at https://github.com/sunyc123r/ARD-SR."
},
{
"venue": "RecSys",
"title": "CRS Arena: Crowdsourced Benchmarking of Conversational Recommender Systems",
"authors": [
"Nolwenn Bernard",
"Hideaki Joko",
"Faegheh Hasibi",
"Krisztian Balog"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2412.10514",
"source": "openalex",
"doi": "https://doi.org/10.1145/3701551.3704120",
"abstract": "We introduce CRS Arena, a research platform for scalable benchmarking of Conversational Recommender Systems (CRS) based on human feedback. The platform displays pairwise battles between anonymous conversational recommender systems, where users interact with the systems one after the other before declaring either a winner or a draw. CRS Arena collects conversations and user feedback, providing a foundation for reliable evaluation and ranking of CRSs. We conduct experiments with CRS Arena on both open and closed crowdsourcing platforms, confirming that both setups produce highly correlated rankings of CRSs and conversations with similar characteristics. We release CRSArena-Dial, a dataset of 474 conversations and their corresponding user feedback, along with a preliminary ranking of the systems based on the Elo rating system. The platform is accessible at https://iai-group-crsarena.hf.space/."
},
{
"venue": "RecSys",
"title": "A Lightweight Method for Modeling Confidence in Recommendations with Learned Beta Distributions",
"authors": [
"Norman Knyazev",
"Harrie Oosterhuis"
],
"year": 2025,
"pdf_url": "https://doi.org/10.5281/zenodo.17352352",
"source": "openalex",
"doi": "https://doi.org/10.5281/zenodo.17352352",
"abstract": "A Lightweight Method for Modeling Confidence in Recommendations with Learned Beta Distributions This repository contains the code used for the experiments in \"A Lightweight Method for Modeling Confidence in Recommendations with Learned Beta Distributions\" published at RecSys 2023 (open access article). Citation If you use this code to produce results for your scientific publication, or if you share a copy or fork, please refer to our RecSys 2023 paper: @inproceedings{knyazev2023alightweight, Author = {Knyazev, Norman and Oosterhuis, Harrie}, Booktitle = {Seventeenth ACM Conference on Recommender Systems (RecSys '23)}, Organization = {ACM}, Title = {A Lightweight Method for Modeling Confidence in Recommendations with Learned Beta Distributions}, Year = {2023} } License The contents of this repository are licensed under the MIT license. If you modify its contents in any way, please link back to this repository. Usage This code makes use of Python 3 and the following packages: jupyter, matplotlib, numpy, scipy, pandas, tqdm, dotenv, tensorflow==2.12.0 and tensorflow-probability. Make sure they are installed. The code can be accessed by running jupyter notebook . in the project folder and navigating to src/notebooks. The process to replicate the results reported in the publication consists of four steps: Modify the variable PROJECT_ROOT in the .env file contained in the root directory of this project to point to the global path of the root directory. Run src/notebooks/preprocess_data.ipynb to download and preprocess the dataset used for evaluation. Run src/notebooks/run_models.ipynb to train the models and export test fold predictions. Each cell trains one model on every one of 10 train-test splits and for each run exports the test set predictions (and the intermediate representations) to logs/LBD_results/{model_name}/{model_name}-{fold_id}-0/export. Run src/notebooks/RQ{research_question_number} to load the above predictions and to obtain the reported numerical results and/or visualizations. Useful tips: By default, training and evaluating different models on multiple folds in src/notebooks/run_models.ipynbis done in a sequential manner. It is also possible to train only some of the models within each runtime by running the chosen cells. Alternatively, all runs for one model can be executed in parallel by setting JOB_TYPE=\"new_process\" or via slurm by setting JOB_TYPE=\"slurm\". For the latter, ensure that src/modules/utils/slurm/slurm_header.txt corresponds to your slurm environment. To evaluate a model on a subset of test folds (e.g. 1), the folds can be specified in the model's config under data_params['params']['folds_to_use_outer'], for example [0, 2, 3, 9]. A single training-evaluation loop can also be executed by running the function src.modules.training.train_run with appropriate parameters."
},
{
"venue": "RecSys",
"title": "Efficient Recommendation with Millions of Items by Dynamic Pruning of Sub-Item Embeddings",
"authors": [
"Aleksandr V. Petrov",
"Craig Macdonald",
"Nicola Tonellotto"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729963",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729963",
"abstract": "A large item catalogue is a major challenge for deploying modern sequential recommender models, since it makes the memory footprint of the model large and increases inference latency. One promising approach to address this is RecJPQ, which replaces item embeddings with sub-item embeddings. However, slow inference remains problematic because finding the top highest-scored items usually requires scoring all items in the catalogue, which may not be feasible for large catalogues. By adapting dynamic pruning concepts from document retrieval, we propose the RecJPQPrune dynamic pruning algorithm to efficiently find the top highest-scored items without computing the scores of all items in the catalogue. Our RecJPQPrune algorithm is safe-up-to-rank K since it theoretically guarantees that no potentially high-scored item is excluded from the final top K recommendation list, thereby ensuring no impact on effectiveness. Our experiments on two large datasets and three recommendation models demonstrate the efficiency achievable using RecJPQPrune: for instance, on the Tmall dataset with 2.2M items, we can reduce the median model scoring time by 64× compared to the Transformer Default baseline, and 5.3× compared to a recent scoring approach called PQTopK. Overall, this paper demonstrates the effective and efficient inference of Transformer-based recommendation models at catalogue scales not previously reported in the literature. Indeed, our RecJPQPrune algorithm can score 2 million items in under 10 milliseconds without GPUs, and without relying on Approximate Nearest Neighbour (ANN) techniques."
},
{
"venue": "RecSys",
"title": "Generative Recommender with End-to-End Learnable Item Tokenization",
"authors": [
"Enze Liu",
"Bowen Zheng",
"Cheng Ling",
"Lantao Hu",
"Han Li",
"Wayne Xin Zhao"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3729989",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3729989",
"abstract": "Generative recommender systems have gained increasing attention as an innovative approach that directly generates item identifiers for recommendation tasks. Despite their potential, a major challenge is the effective construction of item identifiers that align well with recommender systems. Current approaches often treat item tokenization and generative recommendation training as separate processes, which can lead to suboptimal performance. To overcome this issue, we introduce ETEGRec, a novel End-To-End Generative Recommender that unifies item tokenization and generative recommendation into a cohesive framework. Built on a dual encoder-decoder architecture, ETEGRec consists of an item tokenizer and a generative recommender. To enable synergistic interaction between these components, we propose a recommendation-oriented alignment strategy, which includes two key optimization objectives: sequence-item alignment and preference-semantic alignment. These objectives tightly couple the learning processes of the item tokenizer and the generative recommender, fostering mutual enhancement. Additionally, we develop an alternating optimization technique to ensure stable and efficient end-to-end training of the entire framework. Extensive experiments demonstrate the superior performance of our approach compared to traditional sequential recommendation models and existing generative recommendation baselines. Our code is available at https://github.com/RUCAIBox/ETEGRec."
},
{
"venue": "RecSys",
"title": "Training Green and Sustainable Recommendation Models: Introducing Carbon Footprint Data into Early Stopping Criteria",
"authors": [
"Giuseppe Spillo",
"Allegra De Filippo",
"Emanuele Fontana",
"Michela Milano",
"Giovanni Semeraro"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3699682.3728336",
"source": "openalex",
"doi": "https://doi.org/10.1145/3699682.3728336",
"abstract": "With the growing focus on Green AI, there is an urgent need for algorithms that are designed to minimize their environmental impact while maintaining satisfying performance.In this paper, we introduce a novel early stopping strategy that considers carbon footprint data while training a recommendation algorithm.In particular, during the training phase, our criterion epoch-by-epoch analyzes the improvement in terms of predictive accuracy and compares it to the increase in carbon emissions.Then, we analyze the trade-off between the scores, and when the accuracy improves at a rate that is not favorable, the training is stopped.In the experimental evaluation, we showed that our strategy could significantly reduce the carbon footprint of several state-ofthe-art recommendation models, with a limited decrease in accuracy and fairness.While more work is needed to automatically balance the trade-off between accuracy and emissions, this paper sheds light on the need for more sustainable recommendation models and takes a significant step toward designing green training strategies."
},
{
"venue": "RecSys",
"title": "LLMInit: A Free Lunch from Large Language Models for Selective Initialization of Recommendation",
"authors": [
"Weizhi Zhang",
"Liangwei Yang",
"Wooseong Yang",
"Henry Peng Zou",
"Yuqing Liu",
"Ke Xu",
"Sourav Medya",
"Philip S. Yu"
],
"year": 2025,
"pdf_url": "https://aclanthology.org/2025.emnlp-industry.141.pdf",
"source": "openalex",
"doi": "https://doi.org/10.18653/v1/2025.emnlp-industry.141",
"abstract": "Weizhi Zhang, Liangwei Yang, Wooseong Yang, Henry Peng Zou, Yuqing Liu, Ke Xu, Sourav Medya, Philip S. Yu. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: Industry Track. 2025."
},
{
"venue": "RecSys",
"title": "Personalized Course Recommendation System: A Multi-Model Machine Learning Framework for Academic Success",
"authors": [
"M.S. Islam",
"A. S. M. Sanwar Hosen"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2673-6470/5/2/17/pdf?version=1747921097",
"source": "openalex",
"doi": "https://doi.org/10.3390/digital5020017",
"abstract": "The increasing complexity of academic programs and student needs necessitates personalized, data-driven academic advising. Traditional heuristic-based methods often fail to optimize course selection, leading to inefficient academic planning and delayed graduations. This study introduces a hierarchical multi-model machine learning framework for personalized course recommendations, integrating five predictive models: Success Probability Model (SPM), Course Fit Score Model (CFSM), Prerequisite Fulfillment Model (PFM), Graduation Priority Model (GPM), and Recommended Load Model (RLM). These models operate independently in a local model framework, generating specialized predictions that are synthesized by a global model framework through a meta-function. The meta-function aggregates predictions to compute a final score for each course and ensures recommendations align with student success probabilities, program requirements, and workload constraints. It enforces key constraints, such as prerequisite satisfaction, workload optimization, and program-specific requirements, refining recommendations to be both academically viable and institutionally compliant. The framework demonstrated strong predictive performance, with root mean squared error values of 0.00956, 0.011713, and 0.005406 for SPM, CFSM, and RLM, respectively. Classification models for PFM and GPM also yielded high accuracy, exceeding 99%. Designed for modularity and adaptability, the framework allows for the integration of additional predictive models and fine-tuning of recommendation priorities to suit institutional needs. This scalable solution enhances academic advising efficiency by transforming granular model predictions into personalized, actionable course recommendations, supporting students in making informed academic decisions."
},
{
"venue": "RecSys",
"title": "Criteria-Aware Graph Filtering: Extremely Fast Yet Accurate Multi-Criteria Recommendation",
"authors": [
"J.T. Park",
"Jaemin Yoo",
"Won-Yong Shin"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714799",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714799",
"abstract": "Multi-criteria (MC) recommender systems, which utilize MC rating information for recommendation, are increasingly widespread in various e-commerce domains. However, the MC recommendation using training-based collaborative filtering, requiring consideration of multiple ratings compared to single-criterion counterparts, often poses practical challenges in achieving state-of-the-art performance along with scalable model training. To solve this problem, we propose CA-GF, a training-free MC recommendation method, which is built upon criteria-aware graph filtering for efficient yet accurate MC recommendations. Specifically, first, we construct an item-item similarity graph using an MC user-expansion graph. Next, we design CA-GF composed of the following key components, including 1) criterion-specific graph filtering where the optimal filter for each criterion is found using various types of polynomial low-pass filters and 2) criteria preference-infused aggregation where the smoothed signals from each criterion are aggregated. We demonstrate that CA-GF is (a) efficient: providing computational efficiency, offering an extremely fast runtime of less than 0.2 seconds even on the largest benchmark dataset, (b) accurate: outperforming benchmark MC recommendation methods, achieving substantial accuracy gains up to 24% compared to the best competitor, and (c) interpretable: providing interpretations for the contribution of each criterion to the model prediction based on visualizations."
},
{
"venue": "RecSys",
"title": "From Query to Conscience: The Importance of Information Retrieval in Empowering Socially Responsible Consumerism",
"authors": [
"Frans van der Sluis",
"Leif Azzopardi",
"Florian Meier"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730347",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730347",
"abstract": "Millions of consumers search for products online each day, aiming to find items that meet their needs at an acceptable price. While price and quality are major factors in purchasing decisions, ethical considerations increasingly influence consumer behavior -giving rise to the socially responsible consumer. Insights from a recent survey of over 600 consumers reveal that many barriers to ethical shopping stem from information-seeking challenges, often leading to decisions made under uncertainty. These challenges contribute to the intention-behaviour gap, where consumers' desire to make ethical choices is undermined by limited or inaccessible information and inefficacy of search systems in supporting responsible decision-making. In this perspectives paper, we argue that the field of Information Retrieval (IR) has a critical role to play by empowering consumers to make more informed and more responsible choices. We present three interrelated perspectives: (1) reframing ethical consumption as an information extraction problem aimed at reducing information asymmetries; (2) redefining product search as a complex task requiring interfaces that lower the cost and burden of responsible search; and (3) reimagining search as a process of knowledge calibration that helps consumers bridge gaps in awareness when making purchasing decisions. Taken together, these perspectives outline a path from query to conscience - one where IR systems help transform everyday product searches into opportunities for more ethical and informed choices. We advocate for the development of new and novel IR systems and interfaces that address the intricacies of socially responsible consumerism, and call on the IR community to build technologies that make ethical decisions more informed, convenient, and aligned with economic realities."
},
{
"venue": "RecSys",
"title": "A Lightweight Method for Modeling Confidence in Recommendations with Learned Beta Distributions",
"authors": [
"Norman Knyazev",
"Harrie Oosterhuis"
],
"year": 2025,
"pdf_url": "https://doi.org/10.5281/zenodo.17352353",
"source": "openalex",
"doi": "https://doi.org/10.5281/zenodo.17352353",
"abstract": "A Lightweight Method for Modeling Confidence in Recommendations with Learned Beta Distributions This repository contains the code used for the experiments in \"A Lightweight Method for Modeling Confidence in Recommendations with Learned Beta Distributions\" published at RecSys 2023 (open access article). Citation If you use this code to produce results for your scientific publication, or if you share a copy or fork, please refer to our RecSys 2023 paper: @inproceedings{knyazev2023alightweight, Author = {Knyazev, Norman and Oosterhuis, Harrie}, Booktitle = {Seventeenth ACM Conference on Recommender Systems (RecSys '23)}, Organization = {ACM}, Title = {A Lightweight Method for Modeling Confidence in Recommendations with Learned Beta Distributions}, Year = {2023} } License The contents of this repository are licensed under the MIT license. If you modify its contents in any way, please link back to this repository. Usage This code makes use of Python 3 and the following packages: jupyter, matplotlib, numpy, scipy, pandas, tqdm, dotenv, tensorflow==2.12.0 and tensorflow-probability. Make sure they are installed. The code can be accessed by running jupyter notebook . in the project folder and navigating to src/notebooks. The process to replicate the results reported in the publication consists of four steps: Modify the variable PROJECT_ROOT in the .env file contained in the root directory of this project to point to the global path of the root directory. Run src/notebooks/preprocess_data.ipynb to download and preprocess the dataset used for evaluation. Run src/notebooks/run_models.ipynb to train the models and export test fold predictions. Each cell trains one model on every one of 10 train-test splits and for each run exports the test set predictions (and the intermediate representations) to logs/LBD_results/{model_name}/{model_name}-{fold_id}-0/export. Run src/notebooks/RQ{research_question_number} to load the above predictions and to obtain the reported numerical results and/or visualizations. Useful tips: By default, training and evaluating different models on multiple folds in src/notebooks/run_models.ipynbis done in a sequential manner. It is also possible to train only some of the models within each runtime by running the chosen cells. Alternatively, all runs for one model can be executed in parallel by setting JOB_TYPE=\"new_process\" or via slurm by setting JOB_TYPE=\"slurm\". For the latter, ensure that src/modules/utils/slurm/slurm_header.txt corresponds to your slurm environment. To evaluate a model on a subset of test folds (e.g. 1), the folds can be specified in the model's config under data_params['params']['folds_to_use_outer'], for example [0, 2, 3, 9]. A single training-evaluation loop can also be executed by running the function src.modules.training.train_run with appropriate parameters."
},
{
"venue": "RecSys",
"title": "You Don't Bring Me Flowers: Mitigating Unwanted Recommendations Through Conformal Risk Control",
"authors": [
"Giovanni De Toni",
"Erasmo Purificato",
"Emília Gómez",
"Andrea Passerini",
"Bruno Lepri",
"Cristian Consonni"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3705328.3748054",
"source": "openalex",
"doi": "https://doi.org/10.1145/3705328.3748054",
"abstract": "Recommenders are significantly shaping online information consumption.While effective at personalizing content, these systems increasingly face criticism for propagating irrelevant, unwanted, and even harmful recommendations.Such content degrades user satisfaction and contributes to significant societal issues, including misinformation, radicalization, and erosion of user trust.Although platforms offer mechanisms to mitigate exposure to undesired content, these mechanisms are often insufficiently effective and slow to adapt to users' feedback.This paper introduces an intuitive, modelagnostic, and distribution-free method that uses conformal risk control to provably bound unwanted content in personalized recommendations by leveraging simple binary feedback on items.We also address a limitation of traditional conformal risk control approaches, i.e., the fact that the recommender can provide a smaller set of recommended items, by leveraging implicit feedback on consumed items to expand the recommendation set while ensuring robust risk mitigation.Our experimental evaluation on data coming from a popular online video-sharing platform demonstrates that our approach ensures an effective and controllable reduction of unwanted recommendations with minimal effort.The source code is available here: https://github.com/geektoni/mitigating-harm-recsys."
},
{
"venue": "RecSys",
"title": "Leveraging ChatGPT to Empower Training-free Dataset Condensation for Content-based Recommendation",
"authors": [
"Jiahao Wu",
"Qijiong Liu",
"Hengchang Hu",
"Wenqi Fan",
"Shengcai Liu",
"Qing Li",
"Xiao-Ming Wu",
"Ke Tang"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3701716.3715555",
"source": "openalex",
"doi": "https://doi.org/10.1145/3701716.3715555",
"abstract": "Modern Content-Based Recommendation (CBR) techniques utilize item content to deliver personalized services, effectively mitigating information overload. However, these methods often require resource-intensive training on large datasets. To address this issue, we explore dataset condensation for textual CBR in this paper. Dataset condensation aims to synthesize a compact yet informative dataset, enabling models to achieve performance comparable to those trained on full datasets. Applying existing approaches to CBR presents two key challenges: (1) the difficulty of synthesizing discrete texts and (2) the inability to preserve user-item preference information. To overcome these limitations, we propose TF-DCon, an efficient dataset condensation method for CBR. TF-DCon employs a prompt-evolution module to guide ChatGPT in condensing discrete texts and integrates a clustering-based module to condense user preferences effectively. Extensive experiments conducted on three real-world datasets demonstrate TF-DCon's effectiveness. Notably, we are able to approximate up to 97% of the original performance while reducing the dataset size by 95% (i.e., dataset MIND). We have released our code and data for other researchers to reproduce our results."
},
{
"venue": "RecSys",
"title": "Collaboration of Large Language Models and Small Recommendation Models for Device-Cloud Recommendation",
"authors": [
"Zheqi Lv",
"Tianyu Zhan",
"Wenjie Wang",
"Xinyu Lin",
"Shengyu Zhang",
"Wenqiao Zhang",
"Jiwei Li",
"Kun Kuang",
"Fei Wu"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2501.05647",
"source": "openalex",
"doi": "https://doi.org/10.1145/3690624.3709335",
"abstract": "Large Language Models (LLMs) for Recommendation (LLM4Rec) is a promising research direction that has demonstrated exceptional performance in this field. However, its inability to capture real-time user preferences greatly limits the practical application of LLM4Rec because (i) LLMs are costly to train and infer frequently, and (ii) LLMs struggle to access real-time data (its large number of parameters poses an obstacle to deployment on devices). Fortunately, small recommendation models (SRMs) can effectively supplement these shortcomings of LLM4Rec diagrams by consuming minimal resources for frequent training and inference, and by conveniently accessing real-time data on devices."
},
{
"venue": "RecSys",
"title": "Do We Really Need Specialization? Evaluating Generalist Text Embeddings for Zero-Shot Recommendation and Search",
"authors": [
"Matteo Attimonelli",
"Alessandro Bellis",
"Claudio Pomo",
"Dietmar Jannach",
"Eugenio Di Sciascio",
"Tommaso Di Noia"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3705328.3748040",
"source": "openalex",
"doi": "https://doi.org/10.1145/3705328.3748040",
"abstract": "Pre-trained language models (PLMs) are widely used to derive semantic representations from item metadata in recommendation and search.In sequential recommendation, PLMs enhance ID-based embeddings through textual metadata, while in product search, they align item characteristics with user intent.Recent studies suggest task and domain-specific fine-tuning are needed to improve representational power.This paper challenges this assumption for e-commerce applications, showing that Generalist Text Embedding Models (GTEs), pre-trained on large-scale corpora, can guarantee strong zero-shot performance without specialized adaptation.Our experiments on popular e-commerce benchmarks demonstrate that GTEs outperform traditional and fine-tuned models in both sequential recommendation and product search.We attribute this to a superior representational power, as they distribute features more evenly across the embedding space.Finally, we show that compressing embedding dimensions by focusing on the most informative directions (e.g., via PCA) effectively reduces noise and improves the performance of specialized models.To ensure reproducibility, we provide our repository at https://github.com/sisinflab/GTE-Zero-Shot-Recsys."
},
{
"venue": "RecSys",
"title": "Leave No One Behind: Fairness-Aware Cross-Domain Recommender Systems for Non-Overlapping Users",
"authors": [
"Weixin Chen",
"Yuhan Zhao",
"Li Chen",
"Weike Pan"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3705328.3748082",
"source": "openalex",
"doi": "https://doi.org/10.1145/3705328.3748082",
"abstract": "Cross-domain recommendation (CDR) methods predominantly leverage overlapping users to transfer knowledge from a source domain to a target domain.However, through empirical studies, we uncover a critical bias inherent in these approaches: while overlapping users experience significant enhancements in recommendation quality, non-overlapping users benefit minimally and even face performance degradation.This unfairness may erode user trust, and, consequently, negatively impact business engagement and revenue.To address this issue, we propose a novel solution that generates virtual source-domain users for non-overlapping target-domain users.Our method utilizes a dual attention mechanism to discern similarities between overlapping and non-overlapping users, thereby synthesizing realistic virtual user embeddings.We further introduce a limiter component that ensures the generated virtual users align with real-data distributions while preserving each user's unique characteristics.Notably, our method is model-agnostic and can be seamlessly integrated into any CDR model.Comprehensive experiments conducted on three public datasets with five CDR baselines demonstrate that our method effectively mitigates the CDR nonoverlapping user bias, without loss of overall accuracy.Our code is publicly available at https://github.com/WeixinChen98/VUG."
},
{
"venue": "RecSys",
"title": "Uncovering Cross-Domain Recommendation Ability of Large Language Models",
"authors": [
"Xinyi Liu",
"Ruijie Wang",
"Dachun Sun",
"Dilek Hakkani Tür",
"Tarek Abdelzaher"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3701716.3717850",
"source": "openalex",
"doi": "https://doi.org/10.1145/3701716.3717850",
"abstract": "Cross-Domain Recommendation (CDR) seeks to enhance item retrieval in low-resource domains by transferring knowledge from high-resource domains. While recent advancements in Large Language Models (LLMs) have demonstrated their potential in Recommender Systems (RS), their ability to effectively transfer domain knowledge for improved recommendations remains underexplored. To bridge this gap, we propose LLM4CDR, a novel CDR pipeline that constructs context-aware prompts by leveraging users' purchase history sequences from a source domain along with shared features between source and target domains. Through extensive experiments, we show that LLM4CDR achieves strong performance, particularly when using LLMs with large parameter sizes and when the source and target domains exhibit smaller domain gaps. For instance, incorporating CD & Vinyl purchase history for recommendations in Movies & TV yields a 64.28% MAP@1 improvement. We further investigate how key factors-source domain data, domain gap, prompt design, and LLM size-impact LLM4CDR's effectiveness in CDR tasks. Our results highlight that LLM4CDR excels when leveraging a single, closely related source domain and benefits significantly from larger LLMs. These insights pave the way for future research on LLM-driven cross-domain recommendations."
},
{
"venue": "RecSys",
"title": "Avenues for artificial intelligence in library and information services",
"authors": [
"И. А. Митрошин"
],
"year": 2025,
"pdf_url": "https://ntb.gpntb.ru/jour/article/download/1450/1064",
"source": "openalex",
"doi": "https://doi.org/10.33186/1027-3689-2025-1-120-134",
"abstract": "The author discusses the key concepts of the artificial intelligence, computerized analysis and machine learning. The chatbots СhatGPT, GigaChat, Alisa can be used in the libraries and information centers to assist in translations of foreign publications, article reviewing, etc. The author examines the possibility of integrating chatbot into the websites to render the first assistance to the users, in particular beyond office hours. The author reviews using AI for literature system reviewing and argues that the selection process is more efficient, and time is saved. He demonstrates the AI capabilities to improve the relevance of response to the search queries in the library computerized systems and to develop user personal account services. The latter would enable to generate personalized recommendations for articles, patents and reviews within the subject scope of studies and to select the most relevant materials for publication. For the system proper operation, the author suggests to develop the system for evaluation of the produced materials and services quality. Based on this system and requested materials, the personalized analytical and recommendation system can be generated to identify the lines of further research and development. Despite underdeveloped technologies, impossibility of total replacement of the humans, high implementation costs, etc., the AI methods and algorithms of learning and analysis enable to computerize several information processing operations, to reveal patterns and trends, t o p redict u ser n eeds, w hich l ays t he w ay f or d eveloping a nd i mproving services in the libraries and information centers"
},
{
"venue": "RecSys",
"title": "Comparative analysis of methodologies and approaches in recommender systems utilizing large language models",
"authors": [
"Salma S. Elmoghazy",
"Marwa A. Shouman",
"Hamdy K. Elminir",
"Gamal Eldin I. Selim"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s10462-025-11189-8.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s10462-025-11189-8",
"abstract": "Abstract Recommendation systems are indispensable technologies nowadays, as they enable analysis of the huge amount of information available on the internet, helping consumers to make decisions effectively. Ongoing efforts are essential to further develop and align them with the evolving demands of the modern era. In the last few years, large language models (LLMs) have made a huge leap in natural language processing. This advancement has directed researchers’ efforts towards employing these models in various fields, including recommender systems, to leverage the vast amount of data they were trained on. This paper presents a comparative study of a set of recent methodologies that adapt LLMs to recommendations. Throughout the discussed research work, we come up with the insight that LLMs offer significant benefits due to the amount of knowledge they possess and their powerful ability to represent textual data effectively, making them useful in common recommendation issues like cold-start. Also, the variety of fine-tuning and in-context learning techniques enables adaptation of LLMs to a wide range of recommendation tasks. We discussed issues addressed in the reviewed research work and the solutions proposed to enhance recommendation systems. To provide a clearer understanding, we propose taxonomies to categorize the reviewed work based on underlying techniques, involving the role of LLMs in recommendations, learning paradigms, and system structures. We explore datasets, recommendation- and language-related metrics commonly used in this domain. Finally, we analyzed findings in related work, highlighting possible strengths and limitations of using LLMs in recommender systems."
},
{
"venue": "RecSys",
"title": "A Comparative Study on the Integration of Eye-Tracking in Recommender Systems",
"authors": [
"Osamah M. Al-Omair"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/1424-8220/25/9/2692/pdf?version=1745488864",
"source": "openalex",
"doi": "https://doi.org/10.3390/s25092692",
"abstract": "This study investigated the integration of eye tracking technologies in recommender systems, focusing on their potential to enhance personalization, accuracy, and user engagement. Eye tracking metrics, including fixation duration and gaze patterns, provide a non-intrusive means of capturing real-time user preferences, which can lead to more effective recommendations. Through a comprehensive comparison of current studies, this paper synthesizes findings on the impact of eye tracking across application domains such as e-commerce and media. The results indicate notable improvements in recommendation accuracy with the use of gaze-based feedback. However, limitations persist, including reliance on controlled environments, limited sample diversity, and the high cost of specialized eye tracking equipment. To address these challenges, this paper proposes a structured framework that systematically integrates eye tracking data into real-time recommendation generation. The framework consists of an Eye Tracking Module, a Preferences Module, and a Recommender Module, creating an adaptive recommendation process that continuously refines user preferences based on implicit gaze-based interactions. This novel approach enhances the adaptability of recommender systems by minimizing reliance on static user profiles. Future research directions include the integration of additional behavioral indicators and the development of accessible eye tracking tools to broaden real-world impact. Eye tracking shows substantial promise in advancing recommender systems but requires further refinement to achieve practical, scalable applications across diverse contexts."
},
{
"venue": "RecSys",
"title": "Joint Evaluation of Fairness and Relevance in Recommender Systems with Pareto Frontier",
"authors": [
"Theresia Veronika Rampisela",
"Tuukka Ruotsalo",
"Maria Maistro",
"Christina Lioma"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3696410.3714589",
"source": "openalex",
"doi": "https://doi.org/10.1145/3696410.3714589",
"abstract": "Fairness and relevance are two important aspects of recommender systems (RSs). Typically, they are evaluated either (i) separately by individual measures of fairness and relevance, or (ii) jointly using a single measure that accounts for fairness with respect to relevance. However, approach (i) often does not provide a reliable joint estimate of the goodness of the models, as it has two different best models: one for fairness and another for relevance. Approach (ii) is also problematic because these measures tend to be ad-hoc and do not relate well to traditional relevance measures, like NDCG. Motivated by this, we present a new approach for jointly evaluating fairness and relevance in RSs: Distance to Pareto Frontier (DPFR). Given some user-item interaction data, we compute their Pareto frontier for a pair of existing relevance and fairness measures, and then use the distance from the frontier as a measure of the jointly achievable fairness and relevance. Our approach is modular and intuitive as it can be computed with existing measures. Experiments with 4 RS models, 3 re-ranking strategies, and 6 datasets show that existing metrics have inconsistent associations with our Pareto-optimal solution, making DPFR a more robust and theoretically well-founded joint measure for assessing fairness and relevance. Our code: https://github.com/theresiavr/DPFR-recsys-evaluation"
},
{
"venue": "RecSys",
"title": "Some Things Never Change: Overcoming Persistent Challenges in Children IR",
"authors": [
"Maria Soledad Pera",
"Theo Huibers",
"Emiliana Murgia",
"Monica Landoni"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3726302.3730270",
"source": "openalex",
"doi": "https://doi.org/10.1145/3726302.3730270",
"abstract": "There is a lack of a steady and solid influx of information retrieval (IR) research that has children (as the user group) as the protagonist. Existing work is scattered, conducted by only a few research groups, and often based on small-scale user studies or data that cannot be widely shared. Moreover, much of the current research focuses on specific age ranges and abilities, neglecting the broader spectrum of children's needs. Consequently, the paucity of IR research on how search and recommender systems serve and/or ultimately affect children translates into one of many 'Low-resource environments' in IR. Drawing from the literature and our experience in this area, we highlight key challenges and encourage greater attention from the IR community to address this critical gap."
},
{
"venue": "RecSys",
"title": "A look into how machine learning is reshaping engineering models: the rise of analysis paralysis, optimal yet infeasible solutions, and the inevitable Rashomon paradox",
"authors": [
"M.Z. Naser"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s44379-025-00020-4.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s44379-025-00020-4",
"abstract": "Abstract The widespread acceptance of empirically derived codal provisions and equations in civil engineering stands in stark contrast to the skepticism facing machine learning (ML) models – despite their shared statistical foundations. This paper examines this tension through the lens of structural engineering and explores how integrating ML confronts traditional engineering philosophies and professional identities. While recent efforts have documented how ML enhances predictive accuracy, optimizes designs, and analyzes complex behaviors, one might raise concerns about human intuition's diminishing role and algorithms' interpretability. To showcase this rarely explored front, this paper presents how ML can be successfully integrated into various engineering problems by means of formulation via deduction, induction, and abduction. Then, this paper identifies three principal paradoxes that could arise when adopting ML: analysis paralysis (increased prediction accuracy leading to a reduced understanding of physical mechanisms), infeasible solutions (optimization resulting in unconventional designs that challenge engineering intuition), and the Rashomon effect (where contradictions in explainability methods and physics arise). This paper addresses these paradoxes and argues the need to rethink shifts in engineering methodologies and engineering education and to harmonize traditional principles with ML."
},
{
"venue": "RecSys",
"title": "Active Large Language Model-Based Knowledge Distillation for Session-Based Recommendation",
"authors": [
"Yingpeng Du",
"Zhu Sun",
"Ziyan Wang",
"Haoyan Chua",
"Jie Zhang",
"Yew-Soon Ong"
],
"year": 2025,
"pdf_url": "https://ojs.aaai.org/index.php/AAAI/article/download/33263/35418",
"source": "openalex",
"doi": "https://doi.org/10.1609/aaai.v39i11.33263",
"abstract": "Large language models (LLMs) provide a promising way for accurate session-based recommendation (SBR), but they demand substantial computational time and memory. Knowledge distillation (KD)-based methods can alleviate these issues by transferring the knowledge to a small student, which trains a student based on the predictions of a cumbersome teacher. However, these methods encounter difficulties for LLM-based KD in SBR. 1) It is expensive to make LLMs predict for all instances in KD. 2) LLMs may make ineffective predictions for some instances in KD, e.g., incorrect predictions for hard instances or similar predictions as existing recommenders for easy instances. In this paper, we propose an active LLM-based KD method in SBR, contributing to sustainable AI. To efficiently distill knowledge from LLMs with limited cost, we propose to extract a small proportion of instances predicted by LLMs. Meanwhile, for a more effective distillation, we propose an active learning strategy to extract instances that are as effective as possible for KD from a theoretical view. Specifically, we first formulate gains based on potential effects (e.g., effective, similar, and incorrect predictions by LLMs) and difficulties (e.g., easy or hard to fit) of instances for KD. Then, we propose to maximize the minimal gains of distillation to find the optimal selection policy for active learning, which can largely avoid extracting ineffective instances in KD. Experiments on real-world datasets show that our method significantly outperforms state-of-the-art methods for SBR."
},
{
"venue": "KDD",
"title": "Intrusion Detection in IoT Networks Using Hyperdimensional Computing: A Case Study on the NSL-KDD Dataset",
"authors": [
"Ghazal Ghajari",
"Elaheh Ghajari",
"Hossein Mohammadi",
"Fathi Amsaad"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2503.03037",
"source": "openalex",
"doi": "https://doi.org/10.1109/satc65530.2025.11136959",
"abstract": "The rapid expansion of Internet of Things (IoT) networks has introduced new security challenges, necessitating efficient and reliable methods for intrusion detection. In this study, a detection framework based on hyperdimensional computing (HDC) is proposed to identify and classify network intrusions using the NSL-KDD dataset, a standard benchmark for intrusion detection systems. By leveraging the capabilities of HDC, including high-dimensional representation and efficient computation, the proposed approach effectively distinguishes various attack categories such as DoS, probe, R2L, and U2R, while accurately identifying normal traffic patterns. Comprehensive evaluations demonstrate that the proposed method achieves an accuracy of $99.54 \\%$, significantly outperforming conventional intrusion detection techniques, making it a promising solution for IoT network security. This work emphasizes the critical role of robust and precise intrusion detection in safeguarding IoT systems against evolving cyber threats."
},
{
"venue": "KDD",
"title": "Network Anomaly Detection for IoT Using Hyperdimensional Computing on NSL-KDD",
"authors": [
"Ghazal Ghajari",
"Ashutosh Ghimire",
"Elaheh Ghajari",
"Fathi Amsaad"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2503.03031",
"source": "openalex",
"doi": "https://doi.org/10.1109/satc65530.2025.11136944",
"abstract": "With the rapid growth of IoT devices, ensuring robust network security has become a critical challenge. Traditional intrusion detection systems (IDSs) often face limitations in detecting sophisticated attacks within high-dimensional and complex data environments. This paper presents a novel approach to network anomaly detection using hyperdimensional computing (HDC) techniques, specifically applied to the NSL-KDD dataset. The proposed method leverages the efficiency of HDC in processing large-scale data to identify both known and unknown attack patterns. The model achieved an accuracy of $91.55 \\%$ on the KDDTrain+ subset, outperforming traditional approaches. These comparative evaluations underscore the model’s superior performance, highlighting its potential in advancing anomaly detection for IoT networks and contributing to more secure and intelligent cybersecurity solutions."
},
{
"venue": "KDD",
"title": "A Deterministic Comparison of Classical Machine Learning and Hybrid Deep Representation Models for Intrusion Detection on NSL-KDD and CICIDS2017",
"authors": [
"Miguel Arcos-Argudo",
"Rodolfo Bojorque",
"Andrés Torres"
],
"year": 2025,
"pdf_url": "https://www.preprints.org/frontend/manuscript/97b85a0e95e6503059640cf94ea395cf/download_pub",
"source": "openalex",
"doi": "https://doi.org/10.20944/preprints202511.0425.v1",
"abstract": "Intrusion detection systems (IDS) must balance detection quality with operational transparency. We present a deterministic, leakage-free comparison of three classical classifiers: Naïve Bayes, Logistic Regression, and Linear Discriminant Analysis; and we propose a hybrid pipeline that trains LR on autoencoder embeddings. Experiments use NSL-KDD and CICIDS2017 under two regimes (with/without SMOTE applied only on training data). All preprocessing (one-hot encoding, scaling, and imputation) is fit on the training split; fixed seeds and deterministic TensorFlow settings ensure exact reproducibility. We report a complete metric set—Accuracy, Precision, Recall, F1, AUC, and False Alarm Rate (FAR)—and release a replication package (code, preprocessing artifacts, and saved prediction scores) to regenerate all reported tables and metrics. On NSL-KDD, AE+LR yields the highest AUC (≈0.904) and the strongest F1 among the evaluated models (e.g., 0.7583 with SMOTE), while LDA slightly edges LR on Accuracy/F1. NB attains very high Precision (≈0.98) but low Recall (≈0.24), resulting in the weakest F1 yet a low FAR due to conservative decisions. On CICIDS2017, LR delivers the best Accuracy/F1 (0.9878/0.9752 without SMOTE), with AE+LR close behind; both approach ceiling AUC (≈0.996). SMOTE provides modest gains on NSL-KDD and limited benefits on CICIDS2017. Overall, LR/LDA remain strong, interpretable baselines, while AE+LR improves separability (AUC) without sacrificing a simple, auditable decision layer for practical IDS deployment."
},
{
"venue": "KDD",
"title": "A Deterministic Comparison of Classical Machine Learning and Hybrid Deep Representation Models for Intrusion Detection on NSL-KDD and CICIDS2017",
"authors": [
"Miguel Arcos-Argudo",
"Rodolfo Bojorque",
"Carlos Andrés Torres Soto"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/1999-4893/18/12/749/pdf",
"source": "openalex",
"doi": "https://doi.org/10.3390/a18120749",
"abstract": "Intrusion detection systems (IDSs) must balance detection quality with operational transparency. We present a deterministic, leakage-free comparison of three classical classifiers: Naïve Bayes (NB), Logistic Regression (LR), and Linear Discriminant Analysis (LDA). We also propose a hybrid pipeline that trains LR on Autoencoder embeddings (AE). Experiments use NSL-KDD and CICIDS2017 under two regimes (with/without SMOTE (Synthetic Minority Oversampling Technique) applied only on training data). All preprocessing (one-hot encoding, scaling, and imputation) is fitted on the training split; fixed seeds and deterministic TensorFlow settings ensure exact reproducibility. We report a complete metric set—Accuracy, Precision, Recall, F1, Area Under the Curve (AUC), and False Alarm Rate (FAR)—and release a replication package (code, preprocessing artifacts, and saved prediction scores) to regenerate all reported tables and metrics. On NSL-KDD, AE+LR yields the highest AUC (≈0.904) and the strongest F1 among the evaluated models (e.g., 0.7583 with SMOTE), while LDA slightly edges LR on Accuracy/F1. NB attains very high Precision (≈0.98) but low Recall (≈0.24), resulting in the weakest F1, yet a low FAR due to conservative decisions. On CICIDS2017, LR delivers the best Accuracy/F1 (0.9878/0.9752 without SMOTE), with AE+LR close behind; both approach ceiling AUC (≈0.996). SMOTE provides modest gains on NSL-KDD and limited benefits on CICIDS2017. Overall, LR/LDA remain strong, interpretable baselines, while AE+LR improves separability (AUC) without sacrificing a simple, auditable decision layer for practical IDS deployment."
},
{
"venue": "KDD",
"title": "Mobile application based on KDD to predict high-crime areas and promote sustainability in citizen security in a district of Lima-Perú",
"authors": [
"Hugo Vega-Huerta",
"Javier Vilca Velasquez",
"Nicolas Anicama Espinoza",
"Gisella Luisa Elena Maquen-Niño",
"Luis Guerra-Grados",
"Jorge Pantoja-Collantes",
"Oscar Benito-Pacheco",
"Juán Guillermo",
"Adegundo Cámara-Figueroa",
"Javier Cabrera-Díaz",
"Rubén Gil-Calvo",
"Frida López-Córdova"
],
"year": 2025,
"pdf_url": "https://public-pages-files-2025.frontiersin.org/journals/computer-science/articles/10.3389/fcomp.2025.1585632/pdf",
"source": "openalex",
"doi": "https://doi.org/10.3389/fcomp.2025.1585632",
"abstract": "Metropolitan Lima faces a serious citizen security situation, reflected in high rates of crime and violence in several districts. The development of a mobile application to identify and predict areas of high crime incidence is proposed. Using historical data of criminal incidents and reports registered by users in the application, models capable of predicting the occurrence of crimes in real time are trained. The data mining process follows the KDD methodology, which includes the stages of selection, preprocessing, transformation, data mining, evaluation and knowledge consolidation. Machine learning algorithms, such as Random Forest and Gradient Boosting, were used to make these predictions. Visualization techniques, such as heat maps, were also used to represent crime events and facilitate their understanding by users. The results show an accuracy of 88% for the Random Forest algorithm and 91% for Gradient Boosting in predicting the occurrence of crimes, which demonstrates the effectiveness of machine learning models to improve citizen security in Metropolitan Lima, therefore these findings have significant implications for crime prevention and suggest that the application of these technologies can be fundamental to address security challenges in the city."
},
{
"venue": "KDD",
"title": "An optimized LSTM-based deep learning model for anomaly network intrusion detection",
"authors": [
"Nitu Dash",
"Sujata Chakravarty",
"Amiya Kumar Rath",
"Nimay Chandra Giri",
"Kareem M. AboRas",
"N. Gowtham"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-85248-z.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-85248-z",
"abstract": "The increasing prevalence of network connections is driving a continuous surge in the requirement for network security and safeguarding against cyberattacks. This has triggered the need to develop and implement intrusion detection systems (IDS), one of the key components of network perimeter aimed at thwarting and alleviating the issues presented by network invaders. Over time, intrusion detection systems have been instrumental in identifying network breaches and deviations. Several researchers have recommended the implementation of machine learning approaches in IDSs to counteract the menace posed by network intruders. Nevertheless, most previously recommended IDSs exhibit a notable false alarm rate. To mitigate this challenge, exploring deep learning methodologies emerges as a viable solution, leveraging their demonstrated efficacy across various domains. Hence, this article proposes an optimized Long Short-Term Memory (LSTM) for identifying anomalies in network traffic. The presented model uses three optimization methods, i.e., Particle Swarm Optimization (PSO), JAYA, and Salp Swarm Algorithm (SSA), to optimize the hyperparameters of LSTM. In this study, NSL KDD, CICIDS, and BoT-IoT datasets are taken into consideration. To evaluate the efficacy of the proposed model, several indicators of performance like Accuracy, Precision, Recall, F-score, True Positive Rate (TPR), False Positive Rate (FPR), and Receiver Operating Characteristic curve (ROC) have been chosen. A comparative analysis of PSO-LSTMIDS, JAYA-LSTMIDS, and SSA-LSTMIDS is conducted. The simulation results demonstrate that SSA-LSTMIDS surpasses all the models examined in this study across all three datasets."
},
{
"venue": "KDD",
"title": "Intrusion Detection System",
"authors": [
"Zhenjian Li"
],
"year": 2025,
"pdf_url": "https://doi.org/10.17148/ijarcce.2025.14627",
"source": "openalex",
"doi": "https://doi.org/10.17148/ijarcce.2025.14627",
"abstract": "This project presents the development of an Intrusion Detection System (IDS) using machine learning techniques to identify and classify potential threats in network traffic.Leveraging the NSL-KDD dataset, which provides a refined and widely accepted benchmark for network intrusion detection research, the system is trained to detect various types of attacks such as DoS, probe, R2L, and U2R.The project involves preprocessing the dataset, feature selection, and applying supervised learning algorithms like Decision Trees, Random Forest, and Support Vector Machines to build an accurate classification model.The goal is to enhance network security by enabling early detection of malicious activities and reducing false positive rates, ultimately providing a reliable and scalable solution for real-time threat detection in modern network environments."
},
{
"venue": "KDD",
"title": "Smart deep learning model for enhanced IoT intrusion detection",
"authors": [
"Faisal S. Alsubaei"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-06363-5.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-06363-5",
"abstract": "Growing volumes and sensitivities of information in the growing IoT require strong cybersecurity measures to adequately counter increasingly sophisticated cyberattacks. Machine learning-based anomaly detection has the potential to be a viable solution through abnormal network traffic behavior identification that foretells intrusions. Existing approaches, however, are usually hampered by the inability to effectively counter the sophisticated and evolving nature of such threats, especially in preprocessing optimization and hyperparameter tuning, which typically adopt conventional machine learning and deep learning models. This paper addresses these limitations with large preprocessing steps followed by hyperparameter tuning of machine learning XGBoost and deep learning Sequential Neural Network (OSNN) algorithms through Grid Search for their best values to improve multiclass intrusion detection across varied datasets. These deep models were then augmented with a variety of various filters, kernels, activation functions, and regularization techniques in an attempt to boost them in detecting complex, multiclass intrusion patterns. The proposed system was tested comprehensively on three challenging datasets: NSL-KDD, UNSW-NB15, and CICIDS2017. The optimized XGBoost model worked exceptionally well on the NSL-KDD dataset with very high accuracy (99.93%), F1-score (99.84%), MCC (99.86%), and a very low FPR (0.0004). The optimized SNN model also performed well on the NSL-KDD dataset with an accuracy of 99.0% and an AUC of 1.00. Also, the OSNN model performed very well on UNSW-NB15 dataset with an accuracy of 96.80% and a loss of 0.0777, as well as on the CICIDS-2017 dataset with an accuracy of 99.53% and a loss of 0.0236. This superb performance of the OSNN model can be explained by the careful optimization of hyperparameters like strong activation functions (ReLU, GeLU, LeakyReLU), learning rates, dropout rates, and regularization techniques that enable it to learn intricate intrusion patterns efficiently using various datasets. These results highlight the potential of our proposed method to enhance intrusion detection, system integrity, fraud prevention, and ultimately optimize overall network performance."
},
{
"venue": "KDD",
"title": "Intrusion detection system based on machine learning using least square support vector machine",
"authors": [
"Pratik Waghmode",
"Manideep Kanumuri",
"Hosam El‐Ocla",
"Tanner Boyle"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-95621-7.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-95621-7",
"abstract": "Security solutions in the cyber world are essential for enforcing protection against network vulnerabilities and data exploitation. Unauthorized access or attack can be avoided in critical systems using a comprehensive approach via an effective intrusion detection system (IDS). Traditional intrusion detection techniques are no longer accurate and effective enough to handle the demands of the big data age. Machine learning (ML) methods can be utilized for intrusion detection since the classifier's performance has significantly increased over the past decade. A significant limitation of most ML-based IDSs is that they often generate alerts for false predictions. This is owing to misclassifications that tend to occur more frequently than actual threats. Despite their effectiveness, these conventional ML-based IDSs often face difficulties scaling to meet the demands of big data. The increasing volume and complexity of datasets pose various challenges, such as high dimensionality, multiple data sources, and the need for a dependable infrastructure. Consequently, the accuracy of an ML model likely declines when irrelevant features are included from a vast dataset. In this paper, the exhaustive feature selection algorithm is employed to assess every possible combination of features in a dataset to evaluate its performance. The selection is based on identifying the feature subset with the highest accuracy. Hence, an ML-based complete security solution is introduced for network intrusion detection using the supervised framework. This framework utilizes quantum-inspired least square support vector machine (LS-SVM) classifier. This algorithm is used to enhance the classification accuracy in terms of reducing false predictions while minimizing the training time. The hyperparameters of our model are tuned by utilizing those selected features to maximize the accuracy. The model developed is verified using three different datasets, which have been widely applied to intrusion detection. The model achieves high detection performance, with accuracy values of 99.3% for NSL-KDD, 99.5% for CIC-IDS-2017, and 93.3% for UNSW-NB15. Precision remains at 1.00 for CIC-IDS-2017 and UNSW-NB15, while recall reaches 1.00 for CIC-IDS-2017, 0.99 for NSL-KDD, and 0.98 for UNSW-NB15. F1-scores follow the same trend, reflecting the classifier's robust prediction capabilities. In addition, our model demonstrates competitive testing time efficiency in 2.8 s for NSL-KDD, 1.0s for CIC-IDS-2017, and 2.8s for UNSW-NB15. Also, our model requires the minimum training time for all datasets compared to other models. These results highlight the LS-SVM-based model's suitability for real-time intrusion detection applications."
},
{
"venue": "KDD",
"title": "A lightweight framework to secure IoT devices with limited resources in cloud environments",
"authors": [
"Vivek Kumar Pandey",
"Dinesh Kumar Sahu",
"Shiv Prakash",
"Rajkumar Singh Rathore",
"Pratibha Dixit",
"Iryna Hunko"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-09885-0.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-09885-0",
"abstract": "Billions of IoT devices increasingly function as gateways to cloud infrastructures, making them an inevitable target of cyber threats because of the limited resources and low processing capabilities of IoT devices. This paper proposes a lightweight decision tree-based intrusion detection framework suitable for real-time anomaly detection in a resource-constrained IoT environment. Finally, the model also makes use of a novel leaf-cut feature optimization strategy and tight adaptive cloud edge intelligence to achieve high accuracy while minimizing memory and computation demand. In terms of memory, they also use only 12.5 MB in it and evaluated on benchmark datasets including NSL-KDD and Bot-IoT, it gives an accuracy of 98.2% and 97.9%, respectively, and less than 1% false positives, thereby giving up to 6.8% accuracy over some traditional models such as SVM and Neural Networks and up to 78% less energy. It is deployed on Raspberry Pi nodes and can do real-time inference in less than 1 ms and 1,250 samples/sec. Due to the energy efficient, scalable, and interpretable architecture of the proposed solution, it can be implemented as a security solution for IoT use cases in Smart cities, industrial automation, health care, and autonomous vehicles."
},
{
"venue": "KDD",
"title": "Deep learning for network security: an Attention-CNN-LSTM model for accurate intrusion detection",
"authors": [
"Abdullah Mujawib Alashjaee"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-07706-y.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-07706-y",
"abstract": "Intrusion Detection Systems (IDS) are vital for protecting networks with evolving cyber threats, that comprises malware, denial-of-service attacks, and botnets. Hence, this study proposes a novel hybrid deep learning model, which associates Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), along with a self-attention method to highlight the utmost informative input features. Here, CNN works with extraction of spatial features, LSTM works with modelling of temporal sequence. The proposed model is termed Attention-CNN-LSTM. The proposed model achieves 94.8-97.5% accuracy and significantly improves Matthews Correlation Coefficient (MCC) and F1-score by evaluating on NSL-KDD and Bot-IoT datasets. An ablation study confirms each component contribution, particularly the attention layer, to overall performance gains. The architecture supports real-time inference with sub-35ms latency. The model also shows strong potential for real-time deployment, processing over 1200 records per second; hence, this work is applicable for high-traffic environments."
},
{
"venue": "KDD",
"title": "Advanced intrusion detection in internet of things using graph attention networks",
"authors": [
"Aamir S. Ahanger",
"Sajad M. Khan",
"Faheem Masoodi",
"Ayodeji Olalekan Salau"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-94624-8.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-94624-8",
"abstract": "Internet of Things (IoT) denotes a system of interconnected devices equipped with processors, sensors, and actuators that capture and exchange meaningful data with other smart systems. IoT technology has been successfully applied across various sectors, including agriculture, supply chain management, education, healthcare, traffic control, and utility services. However, the diverse range of IoT nodes introduces significant security challenges. Common Internet of Things safety features like encryption, authentication, and access control frequently fall short of meeting their desired functions. In this paper, we present a novel perspective to IoT security by using a Graph-based (GB) algorithm to construct a graph that is evaluated with a graph-based learning Intrusion Detection System (IDS) incorporating a Graph Attention Network (GAT). In addition, we leveraged a small benchmark NSL-KDD dataset to conduct detailed performance evaluation of the GNN model by focusing on essential key metrics such as F1-score, recall, accuracy, and precision to guarantee comprehensive analysis. Our findings validate the effectiveness of the GNN-based IDS in detecting intrusions, which highlights its robustness and scalability in mitigating the evolving security challenges within IoT systems."
},
{
"venue": "KDD",
"title": "Optimizing Intrusion Detection in Wireless Sensor Networks via the Improved Chameleon Swarm Algorithm for Feature Selection",
"authors": [
"Laith Abualigah",
"Mohammad H. Almomani",
"Saleh Ali Alomari",
"Raed Abu Zitar",
"Hazem Migdady",
"Kashif Saleem",
"Václav Snåšel",
"Aseel Smerat",
"Absalom E. Ezugwu"
],
"year": 2025,
"pdf_url": "https://onlinelibrary.wiley.com/doi/pdfdirect/10.1049/cmu2.70029",
"source": "openalex",
"doi": "https://doi.org/10.1049/cmu2.70029",
"abstract": "ABSTRACT In this paper, the improved chameleon swarm algorithm (ICSA) enhances the exploration–exploitation balance while optimizing feature subset selection. The integration of Lévy flight‐based exploration refines ICSA's search strategy, complemented by rotation‐type refinement and adaptive parameter‐setting mechanisms. These modifications ensure that exploration aligns effectively with the feature selection process, leading to a more adaptive and efficient approach. To evaluate ICSA's effectiveness, it is tested on the NSL‐KDD benchmark, a well‐established dataset in intrusion detection systems. Performance is assessed based on key metrics, including accuracy, detection rate, false alarm rate, execution time, and the number of selected features. Comparative analysis against six advanced classifiers demonstrates that ICSA achieves superior results with minimal computational overhead. The algorithm attains the highest accuracy (97.91%) and detection rate (98.75%), the fastest execution time, and the lowest false alarm rate (0.0021), eliminating the need for excessive feature selection. These results confirm that modifying feature selection mechanisms within ICSA significantly enhances computational efficiency and detection performance, as validated through rigorous experimental testing at the classifier level."
},
{
"venue": "KDD",
"title": "Efficient Two-Stage Intrusion Detection System Based on Hybrid Feature Selection Techniques and Machine Learning Classifiers",
"authors": [],
"year": 2025,
"pdf_url": "https://doi.org/10.22266/ijies2025.0430.16",
"source": "openalex",
"doi": "https://doi.org/10.22266/ijies2025.0430.16",
"abstract": "The growing threats of cyberattacks make the integration of machine learning into Intrusion Detection Systems (IDS) increasingly important.However, existing approaches face challenges in effectively analysing diverse types of attacks within complex network architectures.This paper presents a new two-stage mechanism fusing Cuckoo Search and K-means clustering with machine learning classifiers for enhancing IDS performance.In the first stage, the dataset is partitioned into two clusters, C1 and C2, based on harmony between records such that similar and relevant subsets of data alone are considered.In the second stage, the Cuckoo Search algorithm is executed in high-harmony clusters, iteratively choosing effective features for intrusion detection.Targeted feature selection reduces search space.The efficiency and effectiveness of the proposed mechanism are validated through UNSW-NB15 and NSL-KDD datasets and mirror improvement in feature optimization and intrusion accuracy.The proposed mechanism effectively discards irrelevant features and considers only salient ones, reducing the feature count to 22 from 49 in the case of UNSW-NB15 and 19 from 41 in the case of the NSL-KDD dataset.The proposed model reaches an astounding accuracy of 99.78% in the case of UNSW-NB15 and 98.7% in the case of NSL-KDD datasets.By reducing search space, computational complexity is reduced, and accuracy and processing time are increased.These results exhibit the effectiveness of the proposed mechanism in improving the decision-making efficiency of IDSs in complex sensor networks for performance and security improvement."
},
{
"venue": "KDD",
"title": "A Deep Learning-Based Approach for the Detection of Various Internet of Things Intrusion Attacks Through Optical Networks",
"authors": [
"Nouman Imtiaz",
"Abdul Wahid",
"Syed Zain Ul Abideen",
"Mian Muhammad Kamal",
"Nabila Sehito",
"Salah Ud‐Din Khan",
"Bal S. Virdee",
"Lida Kouhalvandi",
"Mohammad Alibakhshikenari"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2304-6732/12/1/35/pdf?version=1735913276",
"source": "openalex",
"doi": "https://doi.org/10.3390/photonics12010035",
"abstract": "The widespread use of the Internet of Things (IoT) has led to significant breakthroughs in various fields but has also exposed critical vulnerabilities to evolving cybersecurity threats. Current Intrusion Detection Systems (IDSs) often fail to provide real-time detection, scalability, and interpretability, particularly in high-speed optical network environments. This research introduces XIoT, which is a novel explainable IoT attack detection model designed to address these challenges. Leveraging advanced deep learning methods, specifically Convolutional Neural Networks (CNNs), XIoT analyzes spectrogram images transformed from IoT network traffic data to detect subtle and complex attack patterns. Unlike traditional approaches, XIoT emphasizes interpretability by integrating explainable AI mechanisms, enabling cybersecurity analysts to understand and trust its predictions. By offering actionable insights into the factors driving its decision making, XIoT supports informed responses to cyber threats. Furthermore, the model’s architecture leverages the high-speed, low-latency characteristics of optical networks, ensuring the efficient processing of large-scale IoT data streams and supporting real-time detection in diverse IoT ecosystems. Comprehensive experiments on benchmark datasets, including KDD CUP99, UNSW NB15, and Bot-IoT, demonstrate XIoT’s exceptional accuracy rates of 99.34%, 99.61%, and 99.21%, respectively, significantly surpassing existing methods in both accuracy and interpretability. These results highlight XIoT’s capability to enhance IoT security by addressing real-world challenges, ensuring robust, scalable, and interpretable protection for IoT networks against sophisticated cyber threats."
},
{
"venue": "KDD",
"title": "A feature selection-driven machine learning framework for anomaly-based intrusion detection systems",
"authors": [
"Emre Emirmahmutoğlu",
"Yılmaz Atay"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s12083-025-01947-4.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s12083-025-01947-4",
"abstract": "Abstract In light of rapid technological developments, a marked rise in global internet usage has contributed to increased sensitive data flow across networks. This increase leads to the diversification of malicious attacks and makes cyber security requirements more evident. In order to ensure network security, intrusion detection systems stand out as an essential component. Intrusion detection systems detect suspicious and malicious activities over network traffic, allowing network administrators and experts to monitor current threats continuously. In anomaly-based systems, machine learning approaches are applied to identify abnormal attempts in network traffic. This study presents a feature selection framework for anomaly-based attack detection systems by combining machine learning and heuristic algorithms. This proposed study aims to improve the performance of IDSs in terms of both time and attack detection by selecting features with heuristic approaches. In the proposed approach, PSO, FPA, DE feature selection methods and LR, DT, RF, KNN, NB, GB, LDA, QDA, AdaBoost, and NN machine learning algorithms are used to perform anomaly-based comparative analyses on KDDCup99, NSL-KDD, UNSW-NB15, CSE-CIS-IDS2018 datasets. Analyses conducted on these datasets with various features demonstrated that models employing feature selection achieved an approximate two-hundred-percent improvement in time efficiency compared to models that did not utilize feature selection. It has been determined that DE, PSO, and FPA, which are used for feature selection, provide high-accuracy outputs when combined with different classifiers. When the analysis results are assessed according to the specified criteria, the highest F1-Score values achieved are as follows: 0.9972 for the DE method in GB, 0.9969 for the PSO method in GB, and 0.9948 for the FPA method in GB, on the KDD CUP 99 dataset. In the NSL-KDD, used as the second dataset, the DE method achieved a score of 0.9713 in GB, the PSO method reached 0.9112 in DT, and the FPA method obtained 0.9894 in RF, respectively. In the third dataset, UNSW-NB15, the DE method achieved a score of 0.9507 in DT, the PSO method reached 0.9068 in DT, and the FPA method obtained 0.8924 in NN. Finally, in the CSE-CIC-IDS2018 dataset, the highest scores achieved using the RF algorithm were 0.99986 for the DE method, 0.99989 for the PSO method, and 0.99987 for the FPA method, based on feature selection. The obtained results underscore the critical role of dataset generation processes and network traffic dynamics in enhancing the performance of intrusion detection systems. Additionally, the significance of feature selection was highlighted. These findings offer valuable insights and present opportunities for further advancements in future research."
},
{
"venue": "KDD",
"title": "Ensemble of feature augmented convolutional neural network and deep autoencoder for efficient detection of network attacks",
"authors": [
"B. Selvakumar",
"M Sivaanandh",
"K. Muneeswaran",
"B. Lakshmanan"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-88243-6.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-88243-6",
"abstract": "Network traffic must be monitored and analyzed for any abnormal activity in order to detect intrusions and to notify administrators of any attacks. A novel ensemble of deep learning technique is proposed to enhance the efficiency of Packet Flow Classification in Network Intrusion Detection System (NIDS). The proposed work consists of three phases: (i) Feature Augmented Convolutional Neural Network (FA-CNN) (ii) Deep Autoencoder (iii) Ensemble of FA-CNN and Deep Autoencoder. In FA-CNN, CNN is trained with augmented features selected using Mutual Information. The FA-CNN is ensembled with Deep Autoencoder to design the ensemble of the classifier. To assess the stated ensemble model, numerous experiments are conducted on benchmark datasets like NSL-KDD and CICDS2017. The result findings are compared with the recent methodologies to assess the performance of the stated work. The results indicate that the suggested work performs better than the existing works with the overall accuracy of 97% for NSLKDD and 95% for CICIDS2017 dataset. Also, the proposed method improved the detection rate of minority attack classes like U2R in NSLKDD and Hearbleed in CICIDS2017."
},
{
"venue": "KDD",
"title": "One-Class Anomaly Detection for Industrial Applications: A Comparative Survey and Experimental Study",
"authors": [
"Davide Paolini",
"Pierpaolo Dini",
"Ettore Soldaini",
"Sergio Saponara"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2073-431X/14/7/281/pdf?version=1752668702",
"source": "openalex",
"doi": "https://doi.org/10.3390/computers14070281",
"abstract": "This article aims to evaluate the runtime effectiveness of various one-class classification (OCC) techniques for anomaly detection in an industrial scenario reproduced in a laboratory setting. To address the limitations posed by restricted access to proprietary data, the study explores OCC methods that learn solely from legitimate network traffic, without requiring labeled malicious samples. After analyzing major publicly available datasets, such as KDD Cup 1999 and TON-IoT, as well as the most widely used OCC techniques, a lightweight and modular intrusion detection system (IDS) was developed in Python. The system was tested in real time on an experimental platform based on Raspberry Pi, within a simulated client–server environment using the NFSv4 protocol over TCP/UDP. Several OCC models were compared, including One-Class SVM, Autoencoder, VAE, and Isolation Forest. The results showed strong performance in terms of detection accuracy and low latency, with the best outcomes achieved using the UNSW-NB15 dataset. The article concludes with a discussion of additional strategies to enhance the runtime analysis of these algorithms, offering insights into potential future applications and improvement directions."
},
{
"venue": "KDD",
"title": "Advanced cloud intrusion detection framework using graph based features transformers and contrastive learning",
"authors": [
"Vijay Govindarajan",
"Junaid Hussain Muzamal"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-07956-w.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-07956-w",
"abstract": "This paper presents a modular and scalable intrusion detection framework that combines graph-based feature extraction, Transformer-based autoencoding, and contrastive learning to improve detection accuracy in cloud environments. Network flows are modeled as graphs to capture relational patterns among IP addresses and services, and a Graph Neural Network (GNN) is used to extract structured embeddings. These embeddings are refined through a Transformer-based autoencoder to preserve contextual information, while contrastive learning enforces clear class separation during classification. The system is evaluated on NSL-KDD and CIC-IDS2018 datasets under both binary and multi-class scenarios. Experimental results show an average accuracy of 99.97%, with high precision and recall across all attack types, including minority classes such as U2R and R2L. The model achieves low false-positive rates and demonstrates real-time inference performance with modest resource requirements. Key contributions include an interpretable pipeline using SHAP for feature attribution, a strategy for mitigating class imbalance, and validation across datasets with detailed security and generalizability analyses. These results support the practical applicability of the proposed approach in high-throughput, cloud-based network environments."
},
{
"venue": "KDD",
"title": "A simulation-driven computational framework for adaptive energy-efficient optimization in machine learning-based intrusion detection systems",
"authors": [
"Ripal Ranpara",
"Osamah Alsalman",
"Om Prakash Kumar",
"Shobhit K. Patel"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-93254-4.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-93254-4",
"abstract": "This paper presents GreenMU, an innovative proposed novel framework designed to address the two main challenges: energy efficiency as one of the main research components and detection performance in intrusion detection systems. In the proposed research paper study, by integrating advanced machine learning techniques such as random forest classifier and support vector machines classifier with knowledge distillation and adaptive energy-aware optimization, GreenMU achieves a balanced trade-off between computational efficiency and cybersecurity accuracy. The proposed MUGuard algorithm is at the framework's core, which dynamically adjusts computational complexity based on real-time actual energy constraints and the evolving threat landscape. Extensive simulations conducted on the KDD 1999 dataset demonstrate that GreenMU achieves a detection accuracy close to 99%, significantly surpassing standard baseline models while reducing energy consumption by 31%. Furthermore, the framework improves computational efficiency, reducing processing time by 15% and making it highly effective for resource-constrained environments such as IoT and edge computing. This research paper study highlights the potential of green artificial intelligence in advancing cybersecurity, providing a scalable, sustainable, and high-performing solution to modern intrusion detection challenges."
},
{
"venue": "KDD",
"title": "Knowledge Discovery in Databases (KDD) Process in Data Mining",
"authors": [
"Mr.Khedekar Nagesh Haridas"
],
"year": 2025,
"pdf_url": "https://www.ijset.in/wp-content/uploads/IJSET_V13_issue6_390.pdf",
"source": "openalex",
"doi": "https://doi.org/10.5281/zenodo.19885681",
"abstract": "Knowledge Discovery in Databases (KDD) is a systematic process used to extract meaningful patterns and useful knowledge from large datasets. It combines techniques from statistics, machine learning, database systems, and artificial intelligence. The KDD process involves several stages including data selection, cleaning, transformation, data mining, and interpretation. This paper explains the complete KDD process, its steps, applications, advantages, and challenges in detail."
},
{
"venue": "KDD",
"title": "Enhanced Anomaly Detection in IoT Networks Using Deep Autoencoders with Feature Selection Techniques",
"authors": [
"Hamza Rhachi",
"Younes Balboul",
"Anas Bouayad"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/1424-8220/25/10/3150/pdf?version=1747398545",
"source": "openalex",
"doi": "https://doi.org/10.3390/s25103150",
"abstract": "An enormous number of the Internet of Things (IoT) applications and their networks have significantly impacted people's lives in diverse situations. With the increasing adoption of these applications in various sectors, ensuring reliability and security has become a critical concern. Moreover, the network that interconnected IoT devices uses advanced communications norms and technologies to capture and transmit data. Still, these networks are subject to various types of attacks that will lead to the loss of user data. Concurrently, the field of anomaly detection for the Internet of Things (IoT) is experiencing rapid expansion. This expansion requires a thorough analysis of application trends and existing gaps. Furthermore, it is critical in detecting interesting phenomena such as device damage and unknown events. However, this task is tough due to the unpredictable nature of anomalies and the complexity of the environment. This paper offers a technique that uses an autoencoder neural network to identify anomalous network communications in IoT networks. More specifically, we propose and implement a model that uses DAE (deep autoencoder) to detect and classify the network data, with an ANOVA F-Test for the feature selection. The proposed model is validated using the NSL-KDD dataset. Compared to some IoT-based anomaly detection models, the experimental results reveal that the suggested model is more efficient at enhancing the accuracy of detecting malicious data. The simulation results show that it works better, with an overall accuracy rate of 85% and 92% successively for the binary and multi-class classifications."
},
{
"venue": "KDD",
"title": "Twined ensemble framework for network security: integrating Random Forest, AdaBoost, and Gradient Boosting for enhanced intrusion detection",
"authors": [
"C. Kishor Kumar Reddy",
"Pulakurthi Anaghaa Reddy",
"Pulakurthi Satyanarayana Reddy",
"Mohammed Shuaib",
"Shadab Alam",
"Sadaf Ahmad",
"A. Rajaram"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s43926-025-00199-1.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s43926-025-00199-1",
"abstract": "An Intrusion Detection System (IDS) is crucial for safeguarding networks against cyber threats. This research presents a novel Twined Ensemble model, specifically designed to enhance intrusion detection performance using the NSL-KDD dataset. The proposed approach leverages a combination of two algorithms from AdaBoost, Gradient Boosting, or Random Forest for each attack category based on their individual performance. These selected classifiers are then combined using a Soft Voting Ensemble technique, which aggregates their probabilistic outputs to yield more accurate and robust predictions. To address the inherent class imbalance in the NSL-KDD dataset, particularly for underrepresented attacks like U2R and R2L, the Synthetic Minority Oversampling Technique (SMOTE) is employed to generate synthetic examples, thereby improving model generalization for rare classes. The Twined Ensemble model is evaluated using metrics such as Accuracy, Precision, Recall, F1-Score, and Cohen’s Kappa. The model achieves 99.68% accuracy for DoS, R2L, and Probe attacks, and 99.83% for U2R attacks, accompanied by a Cohen’s Kappa score of 1.0, indicating near-perfect classification. This architecture effectively integrates adaptive ensemble learning with class balancing strategies, offering a powerful and reliable solution for modern network intrusion detection. Beyond classification accuracy, this study also evaluates the computational performance of each classifier in terms of training time, prediction latency, and memory consumption. Gradient Boosting, while more accurate, exhibits higher training overhead (428.57 s), whereas AdaBoost and Random Forest maintain significantly faster training times (23.05 s and 24.93 s respectively) with minimal memory usage (< 0.04 MB). These findings demonstrate potential scope for the model’s feasibility for real-time IDS deployment in resource-constrained environments."
},
{
"venue": "KDD",
"title": "Knowledge Discovery in Databases (KDD) Process in Data Mining",
"authors": [
"Mr.Khedekar Nagesh Haridas"
],
"year": 2025,
"pdf_url": "https://doi.org/10.5281/zenodo.19885680",
"source": "openalex",
"doi": "https://doi.org/10.5281/zenodo.19885680",
"abstract": "Knowledge Discovery in Databases (KDD) is a systematic process used to extract meaningful patterns and useful knowledge from large datasets. It combines techniques from statistics, machine learning, database systems, and artificial intelligence. The KDD process involves several stages including data selection, cleaning, transformation, data mining, and interpretation. This paper explains the complete KDD process, its steps, applications, advantages, and challenges in detail."
},
{
"venue": "KDD",
"title": "PCA and PSO based optimized support vector machine for efficient intrusion detection in internet of things",
"authors": [
"Mutkule Prasad Raghunath",
"Shyam Deshmukh",
"Poonam Chaudhari",
"Sunil L. Bangare",
"Kishori Kasat",
"Mohan Awasthy",
"Батырхан Омаров",
"Rajesh R. Waghulde"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.measen.2024.101806",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.measen.2024.101806",
"abstract": "The Internet of Things (IoT) is a network that interconnects many everyday objects, including computers, televisions, washing machines, and even whole urban areas. These devices has the capability to collect and disseminate information because to their integration of electronics, software, sensors, and connectivity to a network. The Internet of Things enables the remote sensing, identification, and control of physical things via the utilisation of existing network infrastructure. By using this function, it becomes feasible to integrate elements of the physical world into computerised systems, resulting in enhanced levels of efficiency, precision, and financial profitability. The Internet of Things (IoT) encompasses a diverse array of applications. The Internet of Things (IoT) may be used in several sectors such as healthcare, smart cities, smart homes, transportation, logistics, agriculture, and smart traffic management. The quantity of Internet of Things (IoT) devices is increasing rapidly and exponentially. The surge in numbers is accompanied by a significant escalation in security vulnerabilities. This article presents the development of an intrusion detection system for the Internet of Things using machine learning and feature selection techniques. The system aims to accurately categorise and forecast attacks on IoT devices. This approach utilises the publicly accessible NSL KDD dataset as its input dataset. During the data collecting process for NSL-KDD, all symbolic qualities are transformed into their corresponding numerical representations. Conversely, all numerical features are translated back into symbolic form at the conclusion of the procedure. Principal component analysis is employed to achieve the objective of attribute extraction. After completing the preparation step, the data set is classified using several machine learning techniques such as support vector machine, linear regression, and random forest. Evaluating the veracity, exactness, and retrieval rate of different machine learning algorithms is crucial for choosing the most effective ones. The accuracy of the Intrusion Detection System (IDS) based on Particle Swarm Optimisation (PSO) is 98.5 percent. The PSO-based SVM method is shown superior performance compared to random forest and linear regression methods in terms of precision, recall, and specificity."
},
{
"venue": "KDD",
"title": "Detecting DDoS threats in IoT-driven 6G-energy hubs networks using machine learning algorithms",
"authors": [
"Hesham A. Sakr",
"Magda I. El-Afifi",
"M. A. El-Mowafy",
"Hegazi Ibrahim"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s42452-025-06716-9.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s42452-025-06716-9",
"abstract": "The advent of 6G networks and the rapid expansion of IoT ecosystems, particularly in energy hub (EH) networks, present significant challenges in ensuring secure and reliable communication. Among these, Distributed Denial-of-Service (DDoS) attacks pose a critical threat to the stability of IoT-enabled infrastructures. This study addresses these challenges by systematically evaluating a range of machine learning models and hybrid ensemble techniques tailored for DDoS detection in IoT-driven 6G EH networks. The performance of Random Forest (RF), Gradient Boosting (GB), Support Vector Machines (SVM), Decision Trees (DT), and K-Nearest Neighbors (KNN), as well as hybrid combinations like RF + KNN, GB + KNN, and GB + DT, was rigorously assessed across three benchmark datasets (CICDDOS2019, KDD-CUP, and UNSW-NB) with varying training data sizes (80% and 60%). Among these, RF + KNN emerged as the most effective, achieving the highest accuracy (99.44%) and F1-score (66.62%) on CICDDOS2019 while maintaining robust performance on UNSW-NB. GB + DT demonstrated superior precision (70.95%) on KDD-CUP, while GB + KNN achieved the highest recall (66.67%) on CICDDOS2019, highlighting its capability to minimize false negatives. The findings emphasize the influence of dataset characteristics and training sizes on model performance. CICDDOS2019 consistently produced the highest accuracy due to its well-defined class separability, while KDD-CUP exhibited greater variability and UNSW-NB provided stable yet moderate precision and recall. Smaller training data sizes generally led to performance degradation in F1-score and recall, although occasional accuracy improvements suggested potential overfitting with larger datasets. This research underscores the importance of tailoring hybrid ensemble models to the specific properties of datasets and attack types, offering a scalable, real-time framework for intrusion detection. By enhancing the security and resilience of IoT-driven 6G EH networks, this study provides practical solutions to mitigate the evolving threats posed by DDoS attacks."
},
{
"venue": "KDD",
"title": "Hybrid lion and exponential PSO-based metaheuristic clustering approach for efficient dynamic data stream management",
"authors": [
"M. Ananthi",
"K Valarmathi",
"A. Ramathilagam",
"RVS Praveen"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-07404-9.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-07404-9",
"abstract": "In dynamic data stream environment, the problem related to the exploration of big data within the real time scenario cannot be addressed through the tracking of each individual historic data even though it is highly memory expensive. Thus, a data stream clustering method is essential for exploring and storing the potential amount of information from the historical data determined in a single pass. The dynamic algorithms developed for clustering need to satisfy the two requirements of concept drift and concept evolution. These dynamic algorithms need to handle the change in the association between the object attributes that are existing within each individual clusters. In this paper, A Hybrid Lion and Exponential PSO-based Metaheuristic Clustering Approach (HLEPSOMCA) is proposed for satisfying the requirements of concept drift and concept evolution during efficient dynamic data stream management. This Metaheuristic Clustering Approach is proposed with the properties of good scalability and minimized number of parameters with respect to the number of clusters and high dimensional data determined from the dataset. It adopted different methods of stochastic optimization and deterministic clustering techniques for centring the clusters in an optimal manner. It further adopted density clustering strategies for determining micro clusters, such that Lion and Exponential PSO can be adopted in the initialization phase for maximizing the performance. The experimental results of this HLEPSOMCA approach with respect to KDD-99 dataset confirmed that the purity achieved by the proposed HLEPSOMCA scheme is improved on an average by 13.24%, better than the bassline approaches used for comparison."
},
{
"venue": "KDD",
"title": "KDD 2025 Workshop on Inference Optimization for Generative AI",
"authors": [
"Panpan Xu",
"Youngsuk Park",
"Lin Lee Cheong",
"Yida Wang",
"Yiying Zhang",
"George Karypis",
"Sherry Marcus"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3711896.3737865",
"source": "openalex",
"doi": "https://doi.org/10.1145/3711896.3737865",
"abstract": "The demand for efficient Large Language Model (LLM) inference has surged with the rising adoption of Generative AI (GenAI) applications, particularly in areas such as agents and retrieval-augmented generation. Efficient inference serves two crucial purposes: it enables the deployment of LLM-centered applications that address critical business needs, while also facilitating rapid experimentation for researchers to extract valuable insights and new understandings. However, despite the field's rapid advancement and interdisciplinary nature, there remains a limited exchange of ideas and methodologies between production-facing practitioners and researchers seeking to experiment with new GenAI concepts quickly. To bridge this gap, we are introducing the first KDD workshop on Inference Optimization for Generative AI. Our goal is to create a collaborative platform where researchers and practitioners working across various use cases and stacks of efficient inference can come together to exchange research ideas, establish connections between different disciplines, and identify challenges and research questions that will shape future work."
},
{
"venue": "KDD",
"title": "Anomaly-based intrusion detection system based on SMOTE-IPF, Whale Optimization Algorithm, and ensemble learning",
"authors": [
"Tibebu Bekele Shana",
"Neetu Kumari",
"Mayank Agarwal",
"Samrat Mondal",
"Upaka Rathnayake"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.iswa.2025.200543",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.iswa.2025.200543",
"abstract": "Nowadays, cybersecurity is a major worldwide problem. Intrusion detection systems (IDS) help guarantee network security by detecting malicious entries from legitimate entries in network traffic data. IDS has considerable potential for detecting dynamic cyber threats, identifying abnormalities, and identifying malicious conduct within the network. In this paper, we propose Machine Learning (ML) models with an emphasis on the Synthetic Minority Over-sampling Technique (SMOTE) with Iterative Partitioning Filter (IPF) for class imbalance and the Whale Optimization Algorithm (WOA) for feature selection. Class imbalance often results in poorly constructed ML models prioritizing the majority class. In addition, the absence of feature selection can lead to higher computational complexity without impacting performance accuracy. This study uses Bagging, AdaBoost, Extreme Gradient Boosting (XGBoost) and Extra Trees Classifier as classification models. The two widely used datasets to assess the proposed method are NLS-KDD and UNSW-NB15. The K-Fold cross-validation technique trains this model to minimize potential overfitting. These models are evaluated based on performance metrics such as accuracy, precision, recall, and F1-score. The experimental results demonstrate that the Extra Trees Classifier significantly outperforms the baseline models and achieves accuracy values of 99.9% for the NSL-KDD dataset and 97% for the UNSW-NB 15 dataset and outperforms all evaluation measures compared to baseline models for multi-classification of the IDS. • Extra Tree Classifier performs the best in comparison to baseline models. • A Multi-classification-based intrusion detection system is implemented. • Whale Optimization Algorithm-based feature selection reduced dimensionality. • NSL-KDD and UNSW-NB15 datasets are used for the evaluation. • Hybrid of SMOTE-IPF is used to overcome the class imbalance."
},
{
"venue": "KDD",
"title": "An efficient data driven framework for intrusion detection in wireless sensor networks using deep learning",
"authors": [
"Priyanshu Sinha",
"Dinesh Kumar Sahu",
"Shiv Prakash",
"Rajkumar Singh Rathore",
"Pratibha Dixit",
"Vivek Pandey",
"Iryna Hunko"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-12867-x.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-12867-x",
"abstract": "Wireless Sensor Networks (WSNs) are considered essential to distributed sensing in agricultural, health and industrial domains. Although WSNs have several advantages, they encounter profound cybersecurity threats owing to their processing capacities and small energy sources. In this research work, an intrusion detection framework based on deep learning is designed: a combination of Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) with an adversarial-aware optimization model. The benchmark datasets of NSL-KDD, CICIDS2017, UNSW-NB15, and CTU-13 are analyzed in a number of ways based on structure, diversity, and deep learning requirements. We propose a compound objective to optimize all of these simultaneously, maximizing detection accuracy, minimizing adversarial vulnerability and ensuring model generalizability. Synthetic oversampling with SMOTE is employed to deal with this. Cross-dataset and intra-dataset experiments are implemented when testing the proposed framework, and it outperforms in terms of robustness and transferability. Our efforts are practical in terms of the deployment of a lightweight and resilient IDS that is suitable for WSN settings."
},
{
"venue": "KDD",
"title": "DB3 Team's Solution For Meta KDD Cup' 25",
"authors": [
"Yikuan Xia",
"Jiazun Chen",
"Yirui Zhan",
"Zhao, Suifeng",
"Weipeng Jiang",
"Chaorui Zhang",
"Wei Han",
"Bo Bai",
"Jun Gao"
],
"year": 2025,
"pdf_url": "https://arxiv.org/pdf/2509.09681",
"source": "openalex",
"doi": "https://doi.org/10.48550/arxiv.2509.09681",
"abstract": "This paper presents the db3 team's winning solution for the Meta CRAG-MM Challenge 2025 at KDD Cup'25. Addressing the challenge's unique multi-modal, multi-turn question answering benchmark (CRAG-MM), we developed a comprehensive framework that integrates tailored retrieval pipelines for different tasks with a unified LLM-tuning approach for hallucination control. Our solution features (1) domain-specific retrieval pipelines handling image-indexed knowledge graphs, web sources, and multi-turn conversations; and (2) advanced refusal training using SFT, DPO, and RL. The system achieved 2nd place in Task 1, 2nd place in Task 2, and 1st place in Task 3, securing the grand prize for excellence in ego-centric queries through superior handling of first-person perspective challenges."
},
{
"venue": "KDD",
"title": "An intelligent hybrid approach combining fuzzy C-means and the sperm whale algorithm for cyber attack detection in IoT networks",
"authors": [
"E. I. Elsedimy",
"Sara M. M. Abohashish"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-024-79230-4.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-024-79230-4",
"abstract": "The Internet of Things (IoTs) has revolutionized cities, enabling them to become smarter. IoTs play an important role in monitoring the traffic cameras, roads, smart farming, connected vehicles, air quality, water level, humidity, and carbon dioxide pollution levels in city buildings. One of the major challenges of smart cities is the cyber threat to sensitive data. This paper presents an intelligent approach for detecting cyberattacks and mitigating malicious events in IoT-based smart systems. The proposed approach, known as FCM-SWA, hybridizes a fuzzy C-mean (FCM) with a sperm whale algorithm (SWA). In the first step, we use a novel SWA optimization algorithm to enhance the FCM performance and provide effective defenses against various types of smart city attacks. Next, we propose an adaptive threshold strategy to enhance the global search capability of SWA and prevent the algorithm from settling into local optima. Lastly, we present an efficient scaling approach that solves the clustering problem and finds the optimal cluster center, striking a balance between exploration and exploration in the search space. The proposed FCM-SWA model does better than related and state-of-the-art methods in terms of accuracy, detection rate, precision rate, and F1-scores, as shown by experiments on the NSL-KDD, AWID, and BoT-IoT datasets."
},
{
"venue": "KDD",
"title": "A New Multi‐Objective Binary Bat Algorithm for Feature Selection in Intrusion Detection Systems",
"authors": [
"Mohamed Amine Laamari",
"Nadjet Kamel"
],
"year": 2025,
"pdf_url": "https://onlinelibrary.wiley.com/doi/pdfdirect/10.1002/cpe.70000",
"source": "openalex",
"doi": "https://doi.org/10.1002/cpe.70000",
"abstract": "ABSTRACT Monitoring network traffic and detecting security threats is a vital task in today's world, and intrusion detection systems (IDS) have become an essential tool for this purpose. However, IDSs have to analyze large volumes of data, which often contain irrelevant and redundant features. This makes the job of IDSs more challenging, as they must sift through all available features to identify attack patterns, leading to longer processing time and reduced detection accuracy. To address this, we propose a new wrapper approach for solving the feature selection (FS) problem. Our proposed approach uses a novel multi‐objective binary bat algorithm (MBBA‐FS) with a decision tree classifier. The MBBA‐FS aims to produce a set of non‐dominated solutions that minimize the number of features used while maintaining a high detection accuracy. Then, we use a frequency ranking method to identify a single subset of relevant features from the resulting set of non‐dominated solutions. We tested the feasibility and performance of our approach against other leading FS methods using various datasets, including KDD CUP 1999, NLS‐KDD, UNSW‐NB15, and several synthetic benchmarks. The experimental results show that MBBA‐FS outperforms existing FS approaches in terms of classification accuracy and number of selected features."
},
{
"venue": "KDD",
"title": "A comparative analysis of DNN-based white-box explainable AI methods in network security",
"authors": [
"Osvaldo Arreche",
"Mustafa Abdallah"
],
"year": 2025,
"pdf_url": "https://jis-eurasipjournals.springeropen.com/counter/pdf/10.1186/s13635-025-00201-x",
"source": "openalex",
"doi": "https://doi.org/10.1186/s13635-025-00201-x",
"abstract": "Abstract New research focuses on creating artificial intelligence (AI) solutions for network intrusion detection systems (NIDS), drawing its inspiration from the ever-growing number of intrusions on networked systems, increasing its complexity and intelligibility. Hence, the use of explainable AI (XAI) techniques in real-world intrusion detection systems comes from the requirement to comprehend and elucidate black-box AI models to security analysts. In an effort to meet such requirements, this paper focuses on applying and evaluating white-box XAI techniques (particularly LRP, IG, and DeepLift) for NIDS via an end-to-end framework for neural network models, using three widely used network intrusion datasets (NSL-KDD, CICIDS-2017, and RoEduNet-SIMARGL2021), assessing its global and local scopes, and examining six distinct assessment measures (descriptive accuracy, sparsity, stability, robustness, efficiency, and completeness). We also compare the performance of white-box XAI methods with black-box XAI methods. The results show that using white-box XAI techniques scores high in robustness and completeness, which are crucial metrics for IDS. Moreover, the source codes for the programs developed for our XAI evaluation framework are available to be improved and used by the research community."
},
{
"venue": "KDD",
"title": "Framework design of Network intrusion detection based on convolutional neural networks",
"authors": [
"Yong Chen"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.procs.2025.05.013",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.procs.2025.05.013",
"abstract": "In view of the limitations of traditional network intrusion detection technologies in dealing with complex attack patterns, this paper proposes a detection framework based on deep learning to improve threat recognition accuracy through automatic feature extraction and nonlinear modeling capabilities. Based on KDD Cup 1999 and UNSW-NB15 data sets, a detection model including convolutional neural network (CNN), long short-term memory network (LSTM) and hybrid architecture was constructed to systematically optimize the data preprocessing process and lightweight model design. Experiments show that the comprehensive performance of the proposed CNN-LSTM fusion model on the two types of data sets is significantly better than that of traditional machine learning and support vector machine models. The detection accuracy of KDD Cup 1999 data set is 90% (F1 value 0.9), and the detection accuracy of UNSW-NB15 data set is 87% (F1 value 0.86). The false positive rate is stable at less than 3%. The experimental results show that the deep learning method significantly enhances the ability to identify covert attacks through multi-level feature abstraction and spatiotemporal correlation modeling, and provides a theoretical basis and technical realization path for building an intelligent network security defense system."
},
{
"venue": "KDD",
"title": "A novel approach to intrusion detection system using hybrid flower pollination and cheetah optimization algorithm",
"authors": [
"Deepshikha Kumari",
"Prashant Pranav",
"Abhinav Sinha",
"Sandip Dutta"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-98296-2.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-98296-2",
"abstract": "The study aims to address critical challenges in network security, particularly the limitations of traditional intrusion detection systems (IDS) in terms of adaptability, detection precision, and high false positive rates in dynamic network environments. A novel hybrid IDS model integrating the Flower Pollination Algorithm (FPA), Cheetah Optimization Algorithm (COA), and Artificial Neural Networks (ANN) is proposed to enhance detection accuracy, reduce false positives, and optimize feature selection, anomaly detection, and rule adaptation. The hybrid FPA-COA-ANN model combines the optimization capabilities of FPA and COA with the predictive power of ANN. The model was evaluated using five benchmark datasets-CICIDS-2017, TII-SSRC, Lu-flow, NSL-KDD, and WSN-DS. Key performance metrics were analysed to assess the model's effectiveness in detecting malicious activities in complex network traffic patterns. The hybrid model demonstrated superior performance compared to existing IDS approaches. It achieved accuracy rates of 0.99 on CICIDS-2017, 1.00 on TII-SSRC, 1.00 on Lu-flow, 0.99 on NSL-KDD, and 0.93 on WSN-DS. The results highlight significant improvements in detection precision and adaptability, alongside a reduction in false positive rates, showcasing the model's robustness and scalability for real-time threat detection. The proposed hybrid FPA-COA-ANN model effectively mitigates the limitations of traditional IDS by offering a robust, scalable, and efficient solution for real-time network threat detection. Its high accuracy and adaptability across diverse benchmark datasets underscore its potential as a critical tool for enhancing cybersecurity defences in dynamic and complex environments."
},
{
"venue": "KDD",
"title": "CLASSIFICATION PROCEDURES FOR INTRUSION DETECTION BASED ON KDD CUP 99 DATA SET",
"authors": [
"Shaker El–Sappagh"
],
"year": 2025,
"pdf_url": "https://papers.ssrn.com/sol3/Delivery.cfm/SSRN_ID3401645_code3121922.pdf?abstractid=3401645&mirid=1",
"source": "openalex",
"doi": "https://doi.org/10.5281/zenodo.17528726",
"abstract": "In network security framework, intrusion detection is one of a benchmark part and is a fundamental way to protect PC from many threads. The huge issue in intrusion detection is presented as a huge number of false alerts; this issue motivates several experts to discover the solution for minifying false alerts according to data mining that is a consideration as analysis procedure utilized in a large data e.g. KDD CUP 99. This paper presented various data mining classification for handling false alerts in intrusion detection as reviewed. According to the result of testing many procedure of data mining on KDD CUP 99 that is no individual procedure can reveal all attack class, with high accuracy and without false alerts. The best accuracy in Multilayer Perceptron is 92%; however, the best Training Time in Rule based model is 4 seconds . It is concluded that ,various procedures should be utilized to handle several of network attacks."
},
{
"venue": "KDD",
"title": "An enhanced deep learning framework for intrusion classification enterprise network using multi-branch CNN-attention architecture",
"authors": [
"AmirHossein Biyouki",
"Saman Lotfipour",
"Behnam Haghi"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-34166-1_reference.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-34166-1",
"abstract": "In this work, we propose a deployment-oriented intrusion detection framework for enterprise networks, combining a multi-branch convolutional neural network (CNN) with channel attention and a fine-tuned decision-tree (DT) classifier. Our system offers transparent, human-interpretable rules with minimal inference overhead. We evaluate the proposed model on two public benchmarks: the CIC-IDS2017 dataset, consisting of over 2 million labeled network flows with 80 + features, and the NSL-KDD dataset, containing 125,000 connection records with 41 features. These datasets challenge the model with multiple flow classification tasks, including both known and unknown attack types. Our evaluation shows that the proposed model outperforms strong CNN-based baselines, achieving 99.28% accuracy and 99.30% ROC-AUC on CIC-IDS2017, with a 5.7% improvement over CNN + DT baselines. On NSL-KDD, the model attains a 99.10% accuracy and 0.997 ROC-AUC, marking a 5.7% gain compared to CNN + DT approaches. Furthermore, we report a cross-dataset transfer improvement, with a + 0.97-point increase in macro-F1 score, demonstrating the model's ability to generalize across temporal and dataset shifts. These results underline the system's effectiveness in both classification accuracy and interpretability for real-world enterprise network security deployment."
},
{
"venue": "KDD",
"title": "Cascaded intrusion detection system using machine learning",
"authors": [
"Md. Khabir Uddin Ahamed",
"Abdul Karim"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.sasc.2024.200182",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.sasc.2024.200182",
"abstract": "Cybercrime is becoming an increasing concern these days. In response to the growing cyberthreat, various intrusion detection systems have been developed and proposed to detect anomalies. However, most detection systems suffer from some common issues, such as a high number of false positives that cause regular behaviors to be detected as intrusions, as well as the system’s excessive complexity. Many single classifier models have accuracy issues since they are unable to detect certain anomalies caused by the attack’s polymorphic and zero-day behavior. The signature-based intrusion detection system (SIDS) is unable to identify zero-day intrusions. On the other side, the anomaly-based intrusion detection system (AIDS) generates a significant number of false-positive alarms. In this research, a cascaded intrusion detection system (CIDS) is proposed by combining the one-class support vector machine (OC-SVM)-based AIDS and the decision tree-based SIDS. OC-SVM is used in conjunction with the newly built Distance-Based Intrusion Classification System (DICS). SIDS that use decision trees can discover and classify anomalies. Because OC-SVM is a binary classifier, the intrusion type is determined by DICS.The suggested method aims to detect both popular and well-known zero-day attacks, as well as their type. The CIDS is evaluated using publicly available benchmark datasets, such as the Knowledge Discovery in Databases (KDD) Cup 1999 and the NSL-KDD dataset. The results of the proposed study show that CIDS outperformed both traditional SIDS and AIDS in terms of performance. Both anomalies and their types are detected with high accuracy. • Proposing a new cascaded intrusion detection system (CIDS) that integrates a one-class support vector machine (OC-SVM) anomaly-based IDS (AIDS) with a decision tree-based signature-based IDS (SIDS). This hybrid approach aims to enhance detection accuracy and reduce false positives. • Emphasizing an efficient pre-processing procedure and extensive experimentation. It includes data cleaning, encoding, scaling, and sampling, highlighting the simplicity and effectiveness of the proposed method despite its straightforward approach. • Designing to detect both known and zero-day attacks. OC-SVM is used for its binary classification capability, and the Distance-Based Intrusion Classification System (DICS) is introduced to classify the type of intrusion, addressing the limitation of traditional SIDS in detecting zero-day attacks. • Effectiveness of the proposed system is validated using the KDD Cup 1999 dataset. Results indicate that the CIDS outperforms traditional SIDS and AIDS in terms of accuracy and the ability to detect various types of intrusions. • Highlighting the limitations of single classifier models, such as high false positive rates in anomaly-based systems and the inability of signature-based systems to detect zero-day attacks. By combining OC-SVM and decision tree classifiers, the proposed system aims to address these issues effectively."
},
{
"venue": "KDD",
"title": "Enhanced Network Detection System using NSL-KDD",
"authors": [
"Tanniru Swetha Menon",
"Gondrala Sasank Krishna",
"Lingam Ratna Kumari",
"Chappidi Naga Manikanta"
],
"year": 2025,
"pdf_url": "https://ijsrem.com/download/enhanced-network-detection-system-using-nsl-kdd/?wpdmdl=61318&refresh=6920f5e70f1e41763767783",
"source": "openalex",
"doi": "https://doi.org/10.55041/ijsrem54236",
"abstract": "ABSTRACT Intrusion detection systems must be able to adjust to various attack patterns in various network environments due to the increasing sophistication of cyber threats. When used in real-world situations, traditional systems that were trained on a single dataset frequently have trouble generalizing. This study suggests a hybrid method that uses ensemble techniques to integrate various machine learning models that have been trained on various datasets. We developed two distinct models that captured a variety of attack signatures and common behavior patterns using various network traffic datasets. Improved detection capabilities were obtained by combining these models with majority voting procedures, stacking classifiers, and weighted averaging. While keeping false positive rates low, our experimental implementation showed improved accuracy in identifying known and unknown intrusions. Python was used in the Google Collab environment to develop the system, enabling scalable computation and simple reproducibility. The hybrid approach outperforms individual models in terms of detection accuracy, precision, recall, and F1-score, according to performance evaluation across a number of metrics. A useful framework for creating more robust network intrusion detection systems is provided by this work. Keywords: Cybersecurity Analytics, Model Fusion, Anomaly Detection, Machine Learning Ensemble, Network Security, and Threat Intelligence"
},
{
"venue": "KDD",
"title": "Multihop Intruder Node Detection Scheme (MINDS) for Secured Drones' FANET Communication",
"authors": [
"Simeon Okechukwu Ajakwe",
"Kazeem Lawrence Olabisi",
"Dong‐Seong Kim"
],
"year": 2025,
"pdf_url": "https://onlinelibrary.wiley.com/doi/pdfdirect/10.1049/itr2.70080",
"source": "openalex",
"doi": "https://doi.org/10.1049/itr2.70080",
"abstract": "ABSTRACT Unmanned aerial vehicles (UAVs) are becoming integral to time‐sensitive logistics and intelligent mobility systems due to their flexibility, low deployment cost, and real‐time connectivity. However, their open and dynamic communication environment—typically organized as flying ad hoc networks (FANETs)—makes them highly vulnerable to a wide spectrum of cyber threats. To address this, we propose a novel multihop intrusion node detection scheme (MINDS) powered by an AI‐driven ensemble learning model, X‐CID, optimized for lightweight drone networks. The proposed system integrates a decentralized multi‐hop architecture with intra‐ and inter‐cluster communication validation, enabling real‐time anomaly detection across the physical, communication, and architectural layers of UAV systems. To improve detection performance under resource constraints, feature selection is applied using the Pearson correlation coefficient (PCC), and model hyperparameters are fine‐tuned using randomized search cross‐validation. Trained and evaluated on three benchmark datasets (WSN‐DS, NSL‐KDD, CICIDS2017) covering 24 distinct attack types, X‐CID outperforms traditional models in F1‐score (up to 99.84%), accuracy (up to 99.70%), and achieves low false alarm rates with competitive latency. The proposed approach ensures robust, scalable, and energy‐efficient security for autonomous drone communication, making it suitable for critical missions in logistics, disaster response, and aerial surveillance."
},
{
"venue": "KDD",
"title": "Efficient anomaly detection in tabular cybersecurity data using large language models",
"authors": [
"Xiaoyong Zhao",
"Xingxin Leng",
"Lei Wang",
"Ningning Wang",
"Yanqiong Liu"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-88050-z.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-88050-z",
"abstract": "In cybersecurity, anomaly detection in tabular data is essential for ensuring information security. While traditional machine learning and deep learning methods have shown some success, they continue to face significant challenges in terms of generalization. To address these limitations, this paper presents an innovative method for tabular data anomaly detection based on large language models, called \"Tabular Anomaly Detection via Guided Prompts\" (TAD-GP). This approach utilizes a 7-billion-parameter open-source model and incorporates strategies such as data sample introduction, anomaly type recognition, chain-of-thought reasoning, multi-turn dialogue, and key information reinforcement. Experimental results indicate that the TAD-GP framework improves F1 scores by 79.31%, 97.96%, and 59.09% on the CICIDS2017, KDD Cup 1999, and UNSW-NB15 datasets, respectively. Furthermore, the smaller-scale TAD-GP model outperforms larger models across multiple datasets, demonstrating its practical potential in environments with constrained computational resources and requirements for private deployment. This method addresses a critical gap in research on anomaly detection in cybersecurity, specifically using small-scale open-source models."
},
{
"venue": "KDD",
"title": "Lightweight machine learning framework for efficient DDoS attack detection in IoT networks",
"authors": [
"Mamoona Nawaz",
"Shireen Tahira",
"Dilawar Shah",
"Shujaat Ali",
"Muhammad Tahir"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1038/s41598-025-10092-0",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-10092-0",
"abstract": "The rapid proliferation of Internet of Things (IoT) devices has introduced significant security challenges, with Distributed Denial of Service (DDoS) attacks posing a critical threat to network integrity. Traditional detection methods often rely on computationally intensive models, rendering them unsuitable for resource-constrained IoT environments. To address this limitation, this study proposes a lightweight and scalable machine learning-based DDoS detection framework specifically designed for IoT networks. Utilizing the NSL-KDD dataset, the framework employs an Extra Trees Classifier (ETC) for feature selection, reducing dimensionality while retaining critical attributes. Reduced features were selected to enhance performance and reduce processing cost. Three supervised learning models, Random Forest, Logistic Regression, and Naïve Bayes, were implemented and evaluated based on their detection accuracy, precision, recall, and F1-score. Experimental results demonstrate that the Random Forest model achieves exceptional accuracy (99.88%), precision (99.93%), recall (99.81%), and F1-score (99.87%), outperforming both Logistic Regression (91.61% accuracy) and Naïve Bayes (87.62% accuracy). Furthermore, the proposed framework significantly reduces computational overhead compared to deep learning-based approaches, making it highly suitable for IoT deployments. This research advances IoT security by providing a scalable, efficient, and accurate solution for detecting DDoS attacks, thereby bridging the gap between high-performance requirements and resource limitations in real-world IoT applications."
},
{
"venue": "KDD",
"title": "Proceedings of the ACM SIGKDD Workshop on CyberSecurity and Intelligence Informatics, CSI-KDD. Preface",
"authors": [
"Hsinchun Chen",
"Marc Daciér",
"Marie Francine Moens",
"Gerhard Paaß",
"Christopher C. Yang"
],
"year": 2025,
"pdf_url": "http://publica.fraunhofer.de/documents/N-105128.html",
"source": "openalex",
"doi": "https://doi.org/10.24406/publica-r-362748",
"abstract": "Computer supported communication and infrastructure are integral parts of modern economy. Their security is of incredible importance to a wide variety of practical domains ranging from Internet service providers to the banking industry and e-commerce, from corporate networks to the intelligence community. The CSI-KDD workshop focuses on novel knowledge discovery methods addressing CyberSecurity and intelligence issues as well as innovative applications demonstrating the effectiveness of data mining in solving real-world security problems. The challenge for novel methods originates from the emergence of new types of contents and protocols, and only an integrated view on all modes promises optimal results. Innovative applications are essential as IT-communication as well as computer-supported technical and social infrastructure have an extremely complex structure and require a comprehensive approach to prevent criminal activities."
},
{
"venue": "KDD",
"title": "DeepTransIDS: Transformer-Based Deep learning Model for Detecting DDoS Attacks on 5G NIDD",
"authors": [
"Kumar Harshdeep",
"Konatham Sumalatha",
"Rohit Mathur"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.rineng.2025.104826",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.rineng.2025.104826",
"abstract": "• A deep learning approach is proposed to classify attack types in 5G Networks, emphasizing on Transformer model-based approach. • CNN, RNN, and Transformer model-based approaches are employed to classify attacks in binary class classification scenarios for attack detection. Whereas, multi-class classification is for the identification of attacks. • A dataset generated from a real 5g-network including 1,215,890 records consisting of various types of attacks is discussed • The proposed work shows suitable applicability and accuracy for multi-class classification scenarios. With the rapid deployment of 5G technology, the security of advanced 5G networks is increasingly challenging. Conventional Intrusion Detection Systems (IDS) rely primarily on pre-5G datasets like NSL-KDD and CIC-IDS2017. These datasets cannot address the peculiar challenges that 5G brings, including low latency, large device density, and network slicing. The proposed DeepTransIDS implements a Transformer-based Intrusion Detection System to analyse network traffic in 5G non-IP data delivery scenarios. Unlike traditional IDS approaches that rely on Convolutional Neural Networks, this work uses the self-attention mechanism of Transformers to enhance the classification performance for multi-class network intrusion detection. The proposed model is trained on 5G-NIDD dataset with 1,215,890 network flows that include benign and a different type of malicious traffic. The transformer model achieves 99.79% multi-classification accuracy with better precision, recall, and F1-score, with an increase in accuracy of 0.10% compared to CNN-based IDS models, though the accuracy of RNN-based IDS model is 99.91% the computational time is significantly high. The confusion matrix analysis also confirms the model's ability to accurately identify intricate attack patterns even in the class of imbalance conditions. The findings confirm the dominance of the Transformer model in real-time intrusion detection in dynamic 5G networks."
},
{
"venue": "KDD",
"title": "A Lightweight Intrusion Detection System for IoT and UAV Using Deep Neural Networks with Knowledge Distillation",
"authors": [
"Treepop Wisanwanichthan",
"Mason Thammawichai"
],
"year": 2025,
"pdf_url": "https://doi.org/10.3390/computers14070291",
"source": "openalex",
"doi": "https://doi.org/10.3390/computers14070291",
"abstract": "Deep neural networks (DNNs) are highly effective for intrusion detection systems (IDS) due to their ability to learn complex patterns and detect potential anomalies within the systems. However, their high resource consumption requirements including memory and computation make them difficult to deploy on low-powered platforms. This study explores the possibility of using knowledge distillation (KD) to reduce constraints such as power and hardware consumption and improve real-time inference speed but maintain high detection accuracy in IDS across all attack types. The technique utilizes the transfer of knowledge from DNNs (teacher) models to more lightweight shallow neural network (student) models. KD has been proven to achieve significant parameter reduction (92–95%) and faster inference speed (7–11%) while improving overall detection performance (up to 6.12%). Experimental results on datasets such as NSL-KDD, UNSW-NB15, CIC-IDS2017, IoTID20, and UAV IDS demonstrate DNN with KD’s effectiveness in achieving high accuracy, precision, F1 score, and area under the curve (AUC) metrics. These findings confirm KD’s ability as a potential edge computing strategy for IoT and UAV devices, which are suitable for resource-constrained environments and lead to real-time anomaly detection for next-generation distributed systems."
},
{
"venue": "KDD",
"title": "ADFCNN-BiLSTM: A Deep Neural Network Based on Attention and Deformable Convolution for Network Intrusion Detection",
"authors": [
"Bin Li",
"Jie Li",
"Mingyu Jia"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/1424-8220/25/5/1382/pdf?version=1740399567",
"source": "openalex",
"doi": "https://doi.org/10.3390/s25051382",
"abstract": "Network intrusion detection systems can identify intrusion behavior in a network by analyzing network traffic data. It is challenging to detect a very small proportion of intrusion data from massive network traffic and identify the attack class in intrusion detection tasks. Many existing intrusion detection studies often fail to fully extract the spatial features of network traffic and make reasonable use of temporal features. In this paper, we propose ADFCNN-BiLSTM, a novel deep neural network for network intrusion detection. ADFCNN-BiLSTM uses deformable convolution and an attention mechanism to adaptively extract the spatial features of network traffic data, and it pays attention to the important features from both channel and spatial perspectives. It uses BiLSTM to mine the temporal features from the traffic data and employs the multi-head attention mechanism to allow the network to focus on the time-series information related to suspicious traffic. In addition, ADFCNN-BiLSTM addresses the issue of class imbalance during the training process at both the data level and algorithm level. We evaluated the proposed ADFCNN-BiLSTM on three standard datasets, i.e., NSL-KDD, UNSW-NB15, and CICDDoS2019. The experimental results show that ADFCNN-BiLSTM outperforms the state-of-the-art model in terms of accuracy, detection rate, and false-positive rate."
},
{
"venue": "KDD",
"title": "AE-DTNN: Autoencoder–Dense–Transformer Neural Network Model for Efficient Anomaly-Based Intrusion Detection Systems",
"authors": [
"Hesham Kamal",
"Maggie Mashaly"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2504-4990/7/3/78/pdf?version=1754492009",
"source": "openalex",
"doi": "https://doi.org/10.3390/make7030078",
"abstract": "In this study, we introduce an enhanced hybrid Autoencoder–Dense–Transformer Neural Network (AE-DTNN) model for developing an effective intrusion detection system (IDS) aimed at improving the performance and robustness of threat detection strategies within a rapidly changing and increasingly complex network landscape. The Autoencoder component restructures network traffic data, while a stack of Dense layers performs feature extraction to generate more meaningful representations. The Transformer network then facilitates highly precise and comprehensive classification. Our strategy incorporates adaptive synthetic sampling (ADASYN) for both binary and multi-class classification tasks, complemented by the edited nearest neighbors (ENN) technique and the use of class weights to mitigate class imbalance issues. In experiments conducted on the NF-BoT-IoT-v2 dataset, the AE-DTNN-based IDS achieved outstanding performance, with 99.98% accuracy in binary classification and 98.30% in multi-class classification. On the NSL-KDD dataset, the model reached 98.57% accuracy for binary classification and 97.50% for multi-class classification. Additionally, the model attained 99.92% and 99.78% accuracy in binary and multi-class classification, respectively, on the CSE-CIC-IDS2018 dataset. These results demonstrate the exceptional effectiveness of the proposed model in contrast to conventional approaches, highlighting its strong potential to detect a broad range of network intrusions with high reliability."
},
{
"venue": "KDD",
"title": "Exploring the NSL-KDD Dataset: A Comprehensive Analysis about Intrusion Detection System (IDS)",
"authors": [
"Bechoo Lal",
"Thoudam Basanta Singh",
"M. Nirmala Devi"
],
"year": 2025,
"pdf_url": "https://www.ijsrst.com/index.php/home/article/download/IJSRST251222614/IJSRST251222614",
"source": "openalex",
"doi": "https://doi.org/10.32628/ijsrst251222614",
"abstract": "In this research article the researcher emphasized the Network threats and hazards are evolving at a high-speed rate in recent years. Many mechanisms (such as firewalls, anti-virus, anti-malware, and spam filters) are being used as security tools to protect networks. An intrusion detection system (IDS) is also an effective and powerful network security system to detect unauthorized and abnormal network traffic flow. This article presents a review of the research trends in network-based intrusion detection systems (NIDS), their approaches, and the most common datasets used to evaluate IDS Models. The analysis reported presented in this paper is based on the supervised machine learning approach logistics and XGB- classifier by using NSL-KDD Dataset. The researcher found that logistic classifier given 0.95% accuracy where as XGBooster Classifier gives the 1.00% accuracy , due to the over fitting the researcher used the hyper parameter tuning XGB classifier and got the 0.99% accuracy. The researcher assured that the developed predictive model is more accurate and efficient to detect the intrusion during the data transmission."
},
{
"venue": "KDD",
"title": "Reinforcing Cyber Security: Defensive Machine Learning based Intrusion Detection System for NSL-KDD",
"authors": [
"Ahmed M. Mattar",
"Mohamed Hussein",
"Ahmed H. Eid"
],
"year": 2025,
"pdf_url": "https://doi.org/10.21608/ejmtc.2025.353334.1301",
"source": "openalex",
"doi": "https://doi.org/10.21608/ejmtc.2025.353334.1301",
"abstract": "In the world of digital transformation, intrusion detection has proven to be valuable in protecting the assets of organizations. In this paper, we propose a new machine learning-based technique; Random Forest (RF) to be implemented as intrusion detection system (IDS), to act as defensive frontier for organizations. However, creating an efficient IDS faces a number of challenges, these challenges summarizes in accuracy (mirrored as false positive rate) and training time. Choosing the right machine learning classifier, to work with the right type of network data is important. Detection accuracy can be enhanced by tuning the classifier towards optimal variables. While, training time can be enhanced by correct pre-processing of network data and selecting the features that are most dominant in correlation with the desired output. We examined several machine learning techniques, we applied several data pre-processing steps on NSL-KDD, also, hyper parameter tuning (manipulation) was performed to optimize classifier performance, finally, feature selection techniques were utilized to reduce training time and enhance overall performance. Random Forest has proven to be the most effective machine learning classifier to be used with NSL-KDD, we achieved the highest accuracy of 99.7% and training time of 30.25 second using only 7 features."
},
{
"venue": "KDD",
"title": "Feature Selection pada Dataset NSL-KDD Menggunakan Algoritma Genetic Algorithm untuk Deteksi Serangan Jaringan",
"authors": [
"Freyro Dobry Sianipar",
"Ruth Amelia Vega S. Meliala",
"Yoseph Christian Sitanggang",
"Adidtya Perdana"
],
"year": 2025,
"pdf_url": "https://journal.arteii.or.id/index.php/Neptunus/article/download/1275/1334",
"source": "openalex",
"doi": "https://doi.org/10.61132/neptunus.v3i4.1275",
"abstract": "Information system security faces serious challenges due to increasingly complex cyber attacks. Intrusion Detection Systems (IDS) require efficient approaches to handle high-dimensional data such as the NSL-KDD dataset with 41 features. This study aims to implement the Genetic Algorithm (GA) for feature selection on the NSL-KDD dataset to improve the efficiency and accuracy of network attack detection. The method used is computational experimental research, involving data preprocessing, GA implementation for feature selection, building a classification model using Random Forest, and performance evaluation based on accuracy, precision, recall, F1-score, and computation time. The results show that GA successfully reduced features from 41 to 12 features (70.7% reduction), significantly improving computational efficiency. However, model accuracy slightly decreased from 0.4973 to 0.4951, indicating that while GA is effective for feature selection, the elimination of certain features may reduce classification capability. The implication of this study is that GA can be used as a tool to simplify intrusion detection models, but it should be combined with parameter optimization and data imbalance handling to achieve more optimal performance."
},
{
"venue": "KDD",
"title": "FPE–Transformer: A Feature Positional Encoding-Based Transformer Model for Attack Detection",
"authors": [
"Hande Çavşi Zaim",
"Esra YOLAÇAN"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2076-3417/15/3/1252/pdf?version=1737878642",
"source": "openalex",
"doi": "https://doi.org/10.3390/app15031252",
"abstract": "The increase in cybersecurity threats has made attack detection systems critically important. Traditional deep learning methods often require large amounts of data and struggle to understand relationships between features effectively. With their self-attention mechanism, Transformers excel in modeling complex relationships and long-term dependencies. They are also adaptable to various data types and sources, making them advantageous in large-scale attack detection scenarios. This paper introduces the FPE–Transformer framework, leveraging the strengths of the Transformer architecture. FPE–Transformer incorporates an innovative feature positional encoding mechanism that encodes the positional information of each feature separately, enabling a deeper understanding of feature relationships and more precise attack detection. Additionally, the model includes a ClassificationHead for enhanced accuracy and complex pattern recognition. The framework’s performance was validated using the NSL-KDD and CIC-IDS2017 datasets, demonstrating its superiority over traditional methods in detecting diverse attack types and improving overall performance. This study highlights FPE–Transformer’s innovative approach and ability to address key limitations of traditional deep learning methods, establishing it as a robust solution for modern attack detection challenges."
},
{
"venue": "KDD",
"title": "Enhancing intrusion detection systems: Innovative deep learning approaches using CNN, RNN, DBN and autoencoders for robust network security",
"authors": [
"Y. Avasn Maruthi Kaizar Hossain",
"Zannatul Ferdous",
"T. S. Mohamed Althaf Wahid",
"Md. Torikur Rahman",
"Uttam Kumar Dey",
"Mohammad Moinul Islam"
],
"year": 2025,
"pdf_url": "https://ph.pollub.pl/index.php/acs/article/download/6667/4877",
"source": "openalex",
"doi": "https://doi.org/10.35784/acs_6667",
"abstract": "The increasing sophistication of cyber threats poses significant challenges to network security. This makes effective intrusion detection system (IDS) more important than ever before. Conventional IDS methods, which often rely on signatures or rules it will struggle to keep up with its complex attacks and evolution. This thesis evaluates and analyze the performance of DL algorithms. They include convolutional neural networks (CNN), recurrent neural networks (RNN), deep belief networks (DBN), and Auto-encoder. Using the models, these models are trained and tested only on the NSL-set. KDD data, which is a widely accepted benchmark for evaluating IDS performance. Results show that the proposed deep learning approach significantly outperforms traditional methods, has a higher detection rate, reduce the false positive rate and the ability to identify both known and unknown intrusions. They leverage the strengths of CNN, RNN, DBN, and autoencoders. Doing this research Advances IDS capabilities by providing a robust and adaptable solution to enhance network security."
},
{
"venue": "KDD",
"title": "AegisFormer-IDS: Fast FT-Transformer for Real-Time Intrusion Detection on NSL-KDD",
"authors": [
"Agha Wafa Abbas"
],
"year": 2025,
"pdf_url": "https://doi.org/10.5281/zenodo.18031859",
"source": "openalex",
"doi": "https://doi.org/10.5281/zenodo.18031859",
"abstract": "In the event of insecurity of network infrastructures due to cyber attacks real time intrusiondetectors systems (IDS) play a critical role in ensuring protection to these networksinfrastructures. This paper is the description of AegisFormer-IDS, a lightweight, feature-efficientFeature-Tokenized Transformer (FT-Transformer) network that is best used to perform a fastbinary classification of network traffic as either normal or anomalous. Our work (based on theNSL-KDD data) embraces best practices such as the label encoding of suit categories, thestandardization of features with numeric values, and binary re-labeling (normal vs. attack) tofacilitate movement to a simplified Transformer encoder architecture. The architecture is verycompact, with projection of features, multi-head self-attention layers and classification head thatis fully connected, allowing quick inference without compromising performance. AegisFormerIDS, trained on 10 epochs with the PyTorch framework, Adam optimization, the cross-entropyloss, and a learning rate scheduler, achieves both over 99 and above 0.99 validation accuracy,and F1-score on a 80/20 of train/validation split. Empirical evidence shows that it is better atspeed, accuracy, and resource consumption than standard deep learning models, and can bedeployed in real-time where resource constraints are present. The gap between transformer-basedprogress and scalable high-fidelity network security utilization is bridged in this work, leading toscalable applications of network security."
},
{
"venue": "KDD",
"title": "A GNN-based Novel Approach to Detect Malicious Traffic in Intrusion Detection System",
"authors": [
"Kanika Gupta",
"Prachi",
"N. Chatterjee",
"HIMANSHI HIMANSHI"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.procs.2025.04.576",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.procs.2025.04.576",
"abstract": "Rapid acceleration in Internet of Things (IoT) has tremendously increased the amount of digital information shared over the network. This increase in information escalates the number of network intrusions to gain access to valuable information flowing in the network. This work presents a Graph Neural Network (GNN) based IDS using benchmark dataset (NSL-KDD). The proposed model detects network invasion, assures security of IoT devices and their associated traffic. Initially, the work applies filter-based feature selection techniques to select significant features from NSL-KDD dataset that can easily differentiate malicious traffic from normal network traffic. Thereafter, GNN is applied on these selected features and experimental results demonstrate the prominence of GNN to recognize intrusion by attaining 93.10% accuracy for binary and 89.64% for multi-class intrusion classification. This work also conducts experiments on various machine learning and ensemble algorithms and the result of the presented study highlights the importance of choosing the most efficient algorithm for the development of advanced IDS. The authors have further compared their results with existing solutions for intrusion detection in terms of accuracy, precision and recall and outperformed them."
},
{
"venue": "KDD",
"title": "A Comparative Study of Machine Learning Algorithms for Intrusion Detection Systems using the NSL-KDD Dataset",
"authors": [
"Rulyansyah Permata Putra",
"Amarudin Amarudin"
],
"year": 2025,
"pdf_url": "https://doi.org/10.32520/stmsi.v14i4.5246",
"source": "openalex",
"doi": "https://doi.org/10.32520/stmsi.v14i4.5246",
"abstract": "In today’s digital era, cyberattacks are becoming increasingly complex, rendering traditional rule-based Intrusion Detection Systems (IDS) often ineffective in recognizing new attack patterns. The primary objective of this study is to design and implement a machine learning model for detecting network intrusions efficiently while minimizing latency, through a comparative analysis of several algorithms: Decision Tree, Random Forest, Support Vector Machine (SVM), and Boosting. The research methodology includes the collection of the NSL-KDD dataset, followed by data transformation, cleaning, normalization, and partitioning into training and testing sets. Each algorithm was trained using tuned parameters, and performance was evaluated using metrics such as accuracy, precision, recall, F1-score, and an analysis of training and prediction time. The results indicate that the Boosting algorithm stands out, achieving an accuracy rate of 99.36%. Boosting also demonstrated greater reliability in detecting minority classes, despite requiring longer training times. The application of machine learning methods—particularly Boosting—proves to be an effective approach to enhancing intrusion detection and can serve as a foundation for developing more adaptive and reliable cybersecurity systems."
},
{
"venue": "KDD",
"title": "ABIDS-VEM: leveraging an equilibrium optimizer and data ramification in association with ensemble learning for anomaly-based intrusion detection system",
"authors": [
"Priyanka Verma",
"Donna O’Shea",
"Thomas Newe",
"Nakul Mehta",
"Nitesh Bharot",
"John G. Breslin"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s11227-025-07292-w.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s11227-025-07292-w",
"abstract": "Abstract The convergence of the Internet of Things (IoT) and Industrial Internet of Things (IIoT) within the Industry 4.0 paradigm leverages software-defined networking, multi-cloud architectures, and edge/fog computing to enhance industrial processes. However, this digital transformation introduces significant cybersecurity and privacy vulnerabilities within the complex, data-intensive IoT/IIoT ecosystems. To mitigate these risks, this research proposes a novel Anomaly-based Intrusion Detection System using Voting-based Ensemble Model (ABIDS-VEM) in Industry 4.0 environments. The VEM architecture synergistically combines multiple machine learning algorithms and gradient boosting frameworks, including CatBoost (CB), XGBoost (XGB), LightGBM (LGBM), Logistic Regression (LR), and Random Forest (RF), to enhance the precision and computational efficiency of intrusion detection systems (IDS) in IoT/IIoT contexts. The proposed framework incorporates a data ramification process, in which the data is divided into multiple parts, feature selection process which is optimized through the Equilibrium Optimizer (EO) algorithm, and outlier detection utilizing the Isolation Forest (IF) method. Comprehensive empirical evaluations were conducted using three benchmark datasets: XIIoTID, NSL-KDD, and UNSW-NB15, to validate the efficacy of the proposed system. The model achieves high accuracy across datasets: 98.1476% for XIIoT-ID, an impressive accuracy of 98.9671% for NSL-KDD, and 94.1327% for UNSW-NB15 dataset. These experimental results demonstrate the potential of this approach to significantly enhance the resilience of critical industrial systems and data against evolving cyber threats, thereby supporting the continued evolution of Industry 4.0 technologies and bolstering the security posture of IoT/IIoT ecosystems. This research contributes to the ongoing efforts to secure the rapidly expanding digital industrial landscape, offering a robust solution for detecting and mitigating sophisticated cyberattacks in the increasingly interconnected and data-driven industrial environments of the future."
},
{
"venue": "KDD",
"title": "AegisFormer-IDS: Fast FT-Transformer for Real-Time Intrusion Detection on NSL-KDD",
"authors": [
"Agha Wafa Abbas"
],
"year": 2025,
"pdf_url": "https://doi.org/10.5281/zenodo.18031860",
"source": "openalex",
"doi": "https://doi.org/10.5281/zenodo.18031860",
"abstract": "In the event of insecurity of network infrastructures due to cyber attacks real time intrusiondetectors systems (IDS) play a critical role in ensuring protection to these networksinfrastructures. This paper is the description of AegisFormer-IDS, a lightweight, feature-efficientFeature-Tokenized Transformer (FT-Transformer) network that is best used to perform a fastbinary classification of network traffic as either normal or anomalous. Our work (based on theNSL-KDD data) embraces best practices such as the label encoding of suit categories, thestandardization of features with numeric values, and binary re-labeling (normal vs. attack) tofacilitate movement to a simplified Transformer encoder architecture. The architecture is verycompact, with projection of features, multi-head self-attention layers and classification head thatis fully connected, allowing quick inference without compromising performance. AegisFormerIDS, trained on 10 epochs with the PyTorch framework, Adam optimization, the cross-entropyloss, and a learning rate scheduler, achieves both over 99 and above 0.99 validation accuracy,and F1-score on a 80/20 of train/validation split. Empirical evidence shows that it is better atspeed, accuracy, and resource consumption than standard deep learning models, and can bedeployed in real-time where resource constraints are present. The gap between transformer-basedprogress and scalable high-fidelity network security utilization is bridged in this work, leading toscalable applications of network security."
},
{
"venue": "KDD",
"title": "Building Robust Food Supply Chains through Routing Optimization: A Case Study of KDD in Kuwait",
"authors": [
"Fay Yousef AlMaatouq",
"Fouz Fawaz Jassim Alshehab",
"Yasmeen Khaled Mohammad AlMuhaini",
"Alperen Bal"
],
"year": 2025,
"pdf_url": "https://ieomsociety.org/proceedings/paris2025/410.pdf",
"source": "openalex",
"doi": "https://doi.org/10.46254/eu08.20250410",
"abstract": "This research investigates supply chain resilience through simulation-based optimization, focusing on the Capacitated Vehicle Routing Problem (CVRP) using Kuwait Danish Dairy Company (KDD) as a case study. It aims to enhance logistic robustness against disruptions such as demand spikes, traffic congestion, and regulatory restrictions. Using Google OR-Tools implemented in Python, scenarios representing baseline operations, urban congestion, demand surges, Ramadan operational constraints, and lockdown scenarios were simulated. The CVRP model minimizes total travel cost, considering vehicle capacity and route constraints to maintain cold chain integrity. Results highlighted significant impacts from compounded disruptions; traffic congestion increased total travel distance by 13%, while demand surges stretched operational limits. Ramadan scenarios required precise routing to meet delivery constraints within shorter timeframes. Lockdown scenarios showed reduced operational costs but potential revenue losses from inaccessible nodes. The comparative analysis demonstrated that optimized routing significantly mitigates disruption effects, improving efficiency by up to 20%. Recommendations include adaptive logistics strategies, real-time data integration, diversified supplier networks, and preemptive inventory management. The findings offer practical insights applicable broadly to food supply chains facing similar vulnerabilities in the Gulf region. This research advances the application of CVRP for enhancing supply chain resilience, offering a robust methodology and actionable solutions for managing logistic vulnerabilities. Future research directions include real-time route adjustments, multi-depot scenarios, and environmental sustainability considerations."
},
{
"venue": "KDD",
"title": "A Hybrid CNN-BiLSTM Model with Self-Attention for Network Intrusion Detection: Comparative Evaluation on the NSL-KDD Dataset",
"authors": [
"K. Sujatha"
],
"year": 2025,
"pdf_url": "https://doi.org/10.5281/zenodo.20051149",
"source": "openalex",
"doi": "https://doi.org/10.5281/zenodo.20051149",
"abstract": "Network intrusion detection systems (NIDS) represent a critical line of defence in modern cybersecurity infrastructure, tasked with identifying malicious network activity from high-dimensional, high-velocity traffic data in real time. Conventional signature-based and statistical anomaly detection approaches have demonstrated limited efficacy against zero-day attacks, low-rate flooding attacks, and obfuscated intrusion patterns that exploit temporal dependencies in packet sequences. This paper proposes a hybrid deep learning architecture that combines one-dimensional convolutional neural networks (1D-CNN) for local spatial feature extraction with bidirectional long short-term memory networks (BiLSTM) for sequential temporal modelling, augmented by a self-attention mechanism that dynamically weights the contribution of each time step to the final classification decision. The proposed CNN-BiLSTM-Attention model is trained and evaluated on the NSL-KDD benchmark dataset, a widely used standard for NIDS research that addresses the class imbalance and redundancy limitations of the original KDD Cup 1999 dataset. The model is benchmarked against four baseline classifiers — logistic regression, support vector machine (SVM), random forest, and XGBoost — across four attack categories: Denial of Service (DoS), Probe, Remote-to-Local (R2L), and the benign traffic class. The proposed model achieves an overall classification accuracy of 94.7%, macro-averaged F1-score of 93.8%, and area under the ROC curve (AUC) of 0.987, outperforming all baseline models across all evaluation metrics. Ablation studies confirm that both the BiLSTM and attention components make statistically significant independent contributions to classification performance beyond the CNN baseline alone. The results demonstrate that the CNN-BiLSTM-Attention architecture provides a robust, generalisable framework for multi-class network intrusion detection that is well-suited for deployment in real-time network security monitoring systems."
},
{
"venue": "KDD",
"title": "Leveraging stacking machine learning models and optimization for improved cyberattack detection",
"authors": [
"Neha Pramanick",
"Jimson Mathew",
"Shitharth Selvarajan",
"Mayank Agarwal"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-01052-9.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-01052-9",
"abstract": "The ever-growing number of complex cyber attacks requires the need for high-level intrusion detection systems (IDS). While the available research deals with traditional, hybrid, and ensemble methods for network data analysis, serious challenges are still being met in terms of producing robust and highly accurate detection systems. There are high hurdles in managing high-dimensional network traffic since current methodologies are limited in dealing with imbalanced data issues of minority classes versus the majority and high false positive rate in classification accuracy. This study introduces an innovative framework that directly addresses these persistent challenges through a novel approach to intrusion detection. The proposed method integrates two ML models: J48 and ExtraTreeClassifier for classification. Besides, we propose an improved equilibrium optimizer (EO) approach whereby the previous EO is modified. In this enhanced equilibrium optimizer (EEO), the Fisher score and accuracy score of the K-Nearest Neighbors (KNN) algorithm select the attributes optimally, whereas the synthetic minority oversampling technique combined with iterative partitioning filters (SMOTE-IPF) used to provide class balancing. The KNN technique is also used for data imputation to improve the overall system accuracy. The superior performance of the framework has been validated experimentally on several benchmark datasets, i.e., NSL-KDD, and UNSW-NB15, achieving 99.7% and 98.1% accuracy and F1 score 99.6 and 98.0 respectively. By subjecting the system to a comparative analysis with recent state-of-the-art works, this paper has shown that the proposed methodology yields better improvement in feature selection precision classification accuracy, handling of minority class instance, less demanding storage and computational efficiency."
},
{
"venue": "KDD",
"title": "A novel lightweight deep learning framework using enhanced pelican optimization for efficient cyberattack detection in the Internet of Things environments",
"authors": [
"Yaozhi Chen",
"Yan Guo",
"Yun Gao",
"Baozhong Liu"
],
"year": 2025,
"pdf_url": "https://jeas.springeropen.com/counter/pdf/10.1186/s44147-025-00635-7",
"source": "openalex",
"doi": "https://doi.org/10.1186/s44147-025-00635-7",
"abstract": "Abstract The extensive use of Internet of Things (IoT) technology produces unprecedented connectivity and cyberattack exposure. Recent attack detection tools have poor accuracy, efficiency, and adaptability in the case of IoT systems with scarce resources. To counter these challenges, the current study proposes a hybrid model incorporating an efficient convolutional neural network (CNN) and an enhanced pelican optimization algorithm (EPOA) to detect IoT network attacks. Inspired by how pelicans hunt, EPOA maximizes CNN’s hyperparameters and feature selection for higher accuracy and efficiency in computation. Experimentation with the Bot-IoT, CICIDS2018, and NSL-KDD datasets validates the performance of the proposed EPOA-based deep learning method for cyberattack detection. The model achieves 98.1% accuracy on Bot-IoT, 97.4% on NSL-KDD, and 97.9% on CICIDS2018, outperforming conventional approaches like long short-term memory (LSTM), gated recurrent unit (GRU), support vector machine (SVM), logistic regression (LR), artificial neural network (ANN), and recurrent neural network (RNN). The model also produces a minimum loss value of 0.17, outperforming other approaches with the shortest execution duration. With its efficient design and high detection performance, the proposed approach is highly suitable for continuous IoT cyberattack detection in practical deployment scenarios."
},
{
"venue": "KDD",
"title": "A Hybrid CNN-BiLSTM Model with Self-Attention for Network Intrusion Detection: Comparative Evaluation on the NSL-KDD Dataset",
"authors": [
"K. Sujatha"
],
"year": 2025,
"pdf_url": "https://www.ijaea.com/papers?paper=A+Hybrid+CNN-BiLSTM+Model+with+Self-Attention+for+Network+Intrusion+Detection%3A+Comparative+Evaluation+on+the+NSL-KDD+Dataset",
"source": "openalex",
"doi": "https://doi.org/10.5281/zenodo.20051148",
"abstract": "Network intrusion detection systems (NIDS) represent a critical line of defence in modern cybersecurity infrastructure, tasked with identifying malicious network activity from high-dimensional, high-velocity traffic data in real time. Conventional signature-based and statistical anomaly detection approaches have demonstrated limited efficacy against zero-day attacks, low-rate flooding attacks, and obfuscated intrusion patterns that exploit temporal dependencies in packet sequences. This paper proposes a hybrid deep learning architecture that combines one-dimensional convolutional neural networks (1D-CNN) for local spatial feature extraction with bidirectional long short-term memory networks (BiLSTM) for sequential temporal modelling, augmented by a self-attention mechanism that dynamically weights the contribution of each time step to the final classification decision. The proposed CNN-BiLSTM-Attention model is trained and evaluated on the NSL-KDD benchmark dataset, a widely used standard for NIDS research that addresses the class imbalance and redundancy limitations of the original KDD Cup 1999 dataset. The model is benchmarked against four baseline classifiers — logistic regression, support vector machine (SVM), random forest, and XGBoost — across four attack categories: Denial of Service (DoS), Probe, Remote-to-Local (R2L), and the benign traffic class. The proposed model achieves an overall classification accuracy of 94.7%, macro-averaged F1-score of 93.8%, and area under the ROC curve (AUC) of 0.987, outperforming all baseline models across all evaluation metrics. Ablation studies confirm that both the BiLSTM and attention components make statistically significant independent contributions to classification performance beyond the CNN baseline alone. The results demonstrate that the CNN-BiLSTM-Attention architecture provides a robust, generalisable framework for multi-class network intrusion detection that is well-suited for deployment in real-time network security monitoring systems."
},
{
"venue": "KDD",
"title": "Automatic weed quantification in potato crops based on a modified convolutional neural network using drone images",
"authors": [
"Kevin Vinueza",
"Ana Lucía Sandoval-Pillajo",
"Adriana Giret",
"Diego Trejo",
"Marco Pusdá-Chulde",
"Iván García-Santillán"
],
"year": 2025,
"pdf_url": "https://dm.ageditor.ar/index.php/dm/article/download/194/1108",
"source": "openalex",
"doi": "https://doi.org/10.56294/dm2025194",
"abstract": "Identifying and quantifying weeds is a crucial aspect of agriculture for efficiently controlling them. Weeds compete with the crop for nutrients, minerals, physical space, sunlight, and water, causing problems in crops ranging from low production to economic losses and environmental deterioration of the land. Weed quantification is generally a manual process requiring significant time and precision. Convolutional Neural Networks (CNN) are very common in weed quantification. Thus, the purpose of this research is the adaptation of the ResNeXt50 CNN architecture for semantic segmentation tasks, focused on the automatic quantification of weeds (Broadleaf dock, Dandelion, Kikuyo grass, and other unidentified classes) in potato fields using RGB images acquired by the DJI Mavic 2 Pro drone. The analytical model was trained following the Knowledge Discovery in Databases (KDD) methodology using Python and the TensorFlow-Keras frameworks. The results indicate that the modified ResNeXt50 model presented a mean IoU of 0.7350, a performance comparable to the values reported by other authors considering fewer weed classes. The Student´s t-test and Pearson correlation coefficient were applied to contrast the weed coverage from the model predictions and the ground truth, indicating no statistically significant differences between both measurements in most weed classes."
},
{
"venue": "KDD",
"title": "CGFL: A Robust Federated Learning Approach for Intrusion Detection Systems Based on Data Generation",
"authors": [
"Shu Feng",
"Luhan Gao",
"Leyi Shi"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2076-3417/15/5/2416/pdf?version=1740397270",
"source": "openalex",
"doi": "https://doi.org/10.3390/app15052416",
"abstract": "The implementation of comprehensive security measures is a critical factor in the rapid growth of industrial control networks. Federated Learning has emerged as a viable solution for safeguarding privacy in machine learning. The effectiveness of pattern detection in models is diminished as a result of the difficulty in extracting attack information from extremely large datasets and obtaining an adequate number of examples for specific types of attacks. A robust Federated Learning method, CGFL, is introduced in this study to resolve the challenges presented by data distribution discrepancies and client class imbalance. By employing a data generation strategy to generate balanced datasets for each client, CGFL enhances the global model. It employs a data generator that integrates artificially generated data with the existing data from local clients by employing label correction and data generation techniques. The geometric median aggregation technique was implemented to enhance the security of the aggregation process. The model was simulated and evaluated using the CIC-IDS2017 dataset, NSL-KDD dataset, and CSE-CIC-IDS2018 dataset. The experimental results indicate that CGFL does an effective job of enhancing the accuracy of ICS attack detection in Federated Learning under imbalanced sample conditions."
},
{
"venue": "KDD",
"title": "IDS-IoT: Intrusion Detection System for the Internet of Things Using Enhanced Long-Short Term Memory",
"authors": [
"Gaurav Meena",
"Ajay Indian"
],
"year": 2025,
"pdf_url": "https://ojs.bonviewpress.com/index.php/AIA/article/download/5066/1660",
"source": "openalex",
"doi": "https://doi.org/10.47852/aiabonview52025066",
"abstract": "Network security and intrusion detection have become significant challenges with the emergent inclusion of Internet of Things (IoT) devices across several domains. In this article, we proposed an enhanced long-short term memory (E-LSTM) method for detecting intrusions in IoT-based datasets for the designing of more resilient and competent intrusion detection systems (IDSs) in the dynamic domain of IoT environments, as well as for the thoughtful selection of models appropriate for various dataset characteristics. Four distinct datasets were used in this study: KDD-Cup’99, NSL-KDD, UNSW-NB15, and CICIoT2023. The aim was to estimate and compare the performance across datasets. We provide subtle insights into model behaviors and their capacity to adjust to the particulars of each dataset through rigorous analysis. The proposed enhanced LSTM approach revealed significant differences in precision, recall, accuracy, and F1-score compared to other approaches like AdaBoost, DNN, RNN, and Logistic Regression. It was discovered that, for every dataset, the accuracy rate exceeded 95%. Received: 22 December 2024 | Revised: 15 July 2025 | Accepted: 14 September 2025 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in KDD-Cup’99 dataset at http://kdd.ics.uci.edu/databases/kddcup99/kddcup99.html, NSL-KDD dataset at http://www.unb.ca/cic/datasets/nsl.html, UNSW-NB15 dataset at http://doi.org/10.1109/MilCIS.2015.7348942, and CICIoT2023 dataset at https://doi.org/10.3390/s23135941. Author Contribution Statement Gaurav Meena: Conceptualization, Software, Validation, Formal analysis, Writing – original draft, Writing – review & editing, Supervision, Project administration. Ajay Indian: Methodology, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization."
},
{
"venue": "KDD",
"title": "Anomaly Detection in Edge Computing using Deep Fuzzy Hypersphere Neural Network Learning Model on NSL-KDD Dataset",
"authors": [
"Sonali Jadhav"
],
"year": 2025,
"pdf_url": "https://www.jisem-journal.com/index.php/journal/article/download/5241/2469",
"source": "openalex",
"doi": "https://doi.org/10.52783/jisem.v10i32s.5241",
"abstract": "IoT devices have been extensively utilized on numerous smart applications such as smart city, healthcare, and Industry. Since IoT devices possess tiny computing power and not capable to compute large volumes of data, in spite of the advantages of IoT, it also possesses inherent drawbacks like latency, bandwidth limitation, reliability concerns, and security risks. Edge computing counteracts these drawbacks by processing the data locally and implemented for processing this much huge sensors data on cloud. The Edge will process the data closer to where it is created so that processing may be accelerated and latency can be reduced, again in Edge Computing a variety of irregularities in data generation are generated by the increasing heterogeneity and complexity of edge devices due to their limitations. Anomaly detection is a crucial task in edge computing systems, where identifying unusual or deviant patterns of data is essential to ensuring system security and reliability. An original Deep Fuzzy Hypersphere neural network learning model (DFHNNLM) is proposed in this paper for effective anomaly detection in edge computing tasks. The proposed method outperforms current state-of-the-art for anomaly detection with existing deep learning techniques. Proposed model is suitable for any anomaly dataset like ECG5000, NSL-KDD. According to the experimental results, the DFHNNLM outperforms both deep learning and conventional machine learning methods in anomaly detection, achieving improvements in F1-score, accuracy, precision, and recall."
},
{
"venue": "KDD",
"title": "Deep Learning-Based Intrusion Detection Systems",
"authors": [
"Mahdi Ajdani"
],
"year": 2025,
"pdf_url": "https://www.igi-global.com/ViewTitle.aspx?TitleId=383299&isxn=9798337315720",
"source": "openalex",
"doi": "https://doi.org/10.4018/ijisp.383299",
"abstract": "Given the increasing growth of cyber-attacks, the need for intrusion detection systems (IDS) with higher accuracy and efficiency is critical. This paper presents a novel approach using Generative Adversarial Networks (GANs) for intrusion detection. The proposed model leverages deep learning to extract complex features and uses GANs to generate synthetic data, improving IDS accuracy and efficiency. This approach reduces false positive and negative rates while increasing the accuracy of detecting unknown attacks. Experimental results on the NSL-KDD and CICIDS2017 datasets show 98.2% accuracy, a 1.5% false positive rate, and a 0.8% false negative rate, outperforming conventional methods. These results confirm that GANs can significantly improve the detection and classification of cyber-attacks. The proposed method is an effective solution to enhance cybersecurity and reduce cyber-attack risks, demonstrating significant improvements in IDS and paving the way for future research in this area."
},
{
"venue": "KDD",
"title": "AI-Enhanced Intrusion Detection for UAV Systems: A Taxonomy and Comparative Review",
"authors": [
"Md. Saiful Islam",
"Ashraf Mahmoud",
"Tarek Sheltami"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2504-446X/9/10/682/pdf?version=1759311559",
"source": "openalex",
"doi": "https://doi.org/10.3390/drones9100682",
"abstract": "The diverse usage of Unmanned Aerial Vehicles (UAVs) across commercial, military, and civil domains has significantly heightened the need for robust cybersecurity mechanisms. Given their reliance on wireless communications, real-time control systems, and sensor integration, UAVs are highly susceptible to cyber intrusions that can disrupt missions, compromise data integrity, or cause physical harm. This paper presents a comprehensive literature review of Intrusion Detection Systems (IDSs) that leverage artificial intelligence (AI) to enhance the security of UAV and UAV swarm environments. Through rigorous analysis of recent peer-reviewed publications, we have examined the studies in terms of AI model algorithm, dataset origin, deployment mode: centralized, distributed or federated. The classification also includes the detection strategy: online versus offline. Results show a dominant preference for centralized, supervised learning using standard datasets such as CICIDS2017, NSL-KDD, and KDDCup99, limiting applicability to real UAV operations. Deep learning (DL) methods, particularly Convolutional Neural Networks (CNNs), Long Short-term Memory (LSTM), and Autoencoders (AEs), demonstrate strong detection accuracy, but often under ideal conditions, lacking resilience to zero-day attacks and real-time constraints. Notably, emerging trends point to lightweight IDS models and federated learning frameworks for scalable, privacy-preserving solutions in UAV swarms. This review underscores key research gaps, including the scarcity of real UAV datasets, the absence of standardized benchmarks, and minimal exploration of lightweight detection schemes, offering a foundation for advancing secure UAV systems."
},
{
"venue": "KDD",
"title": "An Efficient Ensemble Network Anomaly Detection System for Cyber-Attacks",
"authors": [
"Saed Alqaraleh"
],
"year": 2025,
"pdf_url": "https://etasr.com/index.php/ETASR/article/download/11920/5334",
"source": "openalex",
"doi": "https://doi.org/10.48084/etasr.11920",
"abstract": "This paper introduces an ensemble-based network anomaly detection system that synergizes classical machine learning classifiers with dimensionality reduction to balance detection accuracy and computational efficiency. The proposed system integrates preprocessing, feature engineering, hybrid learning, and ensemble decision-making to achieve robust anomaly detection and attack classification. Five algorithms, K-Nearest Neighbor (KNN), Naïve Bayes (NB), Random Forest (RF), AdaBoost, and Gradient Boosting (GB), were evaluated both as standalone models and within a soft-voting ensemble framework. To address the high-dimensionality challenges in cybersecurity data, Principal Component Analysis (PCA) was used to retain 95% variance in features while reducing dimensionality by 54% (from 41 to 19 features), achieving a latency improvement of 38% without compromising critical attack detection. A dual-phase SMOTE strategy mitigates class imbalance, enabling 100% recall for rare U2R attacks. Extensive experiments on the KDD CUP99 benchmark demonstrate the superiority of the ensemble method, achieving 93.7% accuracy (vs. 77.7–90% for individual models). Furthermore, while GB achieved the highest individual average performance at 90%, the proposed ensemble exhibited strong performance in adversarial tests, gaining 97.1% accuracy compared to GB's 85.2% against GAN-generated attacks. These findings establish a foundation for adaptive cybersecurity systems that employ machine learning to tackle emerging adversarial defense mechanisms, highlighting accuracy and operational feasibility in evolving threat landscapes."
},
{
"venue": "KDD",
"title": "PERBANDINGAN NAIVE BAYES, SUPPORT VECTOR MACHINE, LOGISTIC REGRESSION DAN RANDOM FOREST DALAM MENGANALISIS SENTIMEN MENGENAI TIKTOKSHOP",
"authors": [
"Octavia Salwa Dzaky Fadhillah",
"Jajam Haerul Jaman",
"Carudin Carudin"
],
"year": 2025,
"pdf_url": "https://journal.eng.unila.ac.id/index.php/jitet/article/download/5746/2314",
"source": "openalex",
"doi": "https://doi.org/10.23960/jitet.v13i1.5746",
"abstract": "Pertumbuhan e-commerce yang pesat di Indonesia dan ramainya pembicaraan salah satu platform yaitu Tiktokshop, mendorong pentingnya analisis sentimen untuk memahami tanggapan publik. Penelitian ini bertujuan menganalisis sentimen pengguna terhadap Tiktokshop melalui tweet di platform X, menggunakan algoritma Naive Bayes, Support Vector Machine (SVM), Logistic Regression, dan Random Forest. Data diambil melalui web scraping dan diproses menggunakan metodologi Knowledge Discovery in Database (KDD). Tahapan KDD meliputi Data Selection, Preprocessing, Transformation, Data Mining, Evaluation, dan Knowledge Presentation. Label sentimen ditentukan dengan pendekatan lexicon, sehingga didapatkan 521 data label negatif dan 502 data label positif. Pengujian performa algoritma klasifikasi menggunakan Confusion Matrix dan Classification Report. Pengujian tersebut menghasilkan nilai akurasi tertinggi pada SVM sebesar 81%, diikuti Random Forest dengan 80%, Logistic Regression dengan 79%, dan Naive Bayes sebesar 75%. Visualisasi word cloud menunjukkan kata-kata dominan untuk sentimen positif seperti ’beli’, ’checkout’, ’barang’, ’murah’, dan ’suka’, sedangkan untuk sentimen negatif yaitu ’belanja’, ’live’, ’habis’ dan ’astaga’. Hasil penelitian ini diharapkan membantu perusahaan dalam mengevaluasi layanan dan strategi pemasaran Tiktokshop."
},
{
"venue": "KDD",
"title": "Deep reinforcement learning-based intrusion detection scheme for software-defined networking",
"authors": [
"R. Kanimozhi",
"P. Ramesh"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-24869-w.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-24869-w",
"abstract": "A robust Deep Reinforcement Learning-based Intrusion Detection Scheme (DRL-IDS) for Software-Defined Networking (SDN) which combines the Long-Short Term Sequence Recurrent Neural Network (LFTS-RNN) with the Particle Cloud-Integrated Joint Time- and Feature-Optimization Algorithm (PC-JTFOA). The hybrid model aims to enhance the security of SDN through the detection and mitigation of a wide array of Distributed Denial of Service attacks and network misbehaviors across different SDN planes. The LFTS-RNN is used for accurate attack detection and misbehavior identification. Meanwhile, the PC-JTFOA optimizes feature selection, load balancing, and energy-efficient routing, thus ensuring fast and reliable network traffic management. The deep reinforcement learning approach further enables continuous adaptation to changing network behaviors, thus making the model dynamically adapt to known as well as emerging attack vectors. The proposed DRL-IDS scheme obtains superior performance in experimental results based on the NSL-KDD and WPPD datasets. The LFTS-RNN model indicates a highly impressive sensitivity of 98.67% and specificity of 97.42%, while the DRL-IDS model presents an detection accuracy of 99.85%. The PC-JTFOA further improves the solution by exhibiting a low response time of 1423 ms, which indicates tremendous improvement in computational efficiency. A comparative analysis with the existent intrusion detection methods pointed out that the scheme proposed not only outperforms other models in terms of detection accuracy as well as adaptability, but it also reduces complexity."
},
{
"venue": "KDD",
"title": "The KDD Process in Big Data Analytics: A Theoretical Approach to Taxpayer Non-Compliance Analysis",
"authors": [
"Arnela Kaknjo",
"Lejla Turulja"
],
"year": 2025,
"pdf_url": "https://doi.org/10.2478/jfap-2025-0002",
"source": "openalex",
"doi": "https://doi.org/10.2478/jfap-2025-0002",
"abstract": "Abstract In the modern business environment, big data analytics and data mining techniques are increasingly recognized as tools for improving fiscal discipline and more efficient management of public revenues. This paper explores the possibility of applying the knowledge discovery process from databases to detect patterns of financial behavior that may indicate tax non-compliance. A quantitative approach based on the analysis of secondary data from ten joint-stock companies from the Federation of Bosnia and Herzegovina, for which financial statements and tax debt data are available, was used. The relationship between key financial indicators (EPS, financial stability ratio, total asset turnover ratio and debt ratio) and the amount of tax debt was examined using descriptive statistics and regression analysis. The results show that lower profitability and poorer financial stability significantly correlate with higher tax debt, while high operational efficiency and debt have a more complex and statistically marginal impact. The findings confirm the possibility of using publicly available financial data for early identification of risky taxpayers, which opens up space for further development of predictive models in the domain of tax analytics."
},
{
"venue": "KDD",
"title": "Knowledge Discovery in Databases (KDD) Applied to the Demand for Technologies for Pasture Conservation and Management System",
"authors": [
"U. G. P. de Abreu",
"P. M. Santos",
"Helano Póvoas de Lima",
"Jayme Garcia Arnal Barbedo",
"Patrícia Belfiore"
],
"year": 2025,
"pdf_url": "https://www.preprints.org/frontend/manuscript/fca3a73a46d98d0a21a7c500b6e3a66b/download_pub",
"source": "openalex",
"doi": "https://doi.org/10.20944/preprints202509.0907.v1",
"abstract": "Most of the ruminant production in Brazil is based on the use of pastures, which are present in every Brazilian biome. The aim of this study was to extract patterns and knowledge from the stakeholders’ response database regarding the technological processes used for pasture conservation and management. Electronic questionnaires were used to perform the stakeholders’ survey, as they are more economical and agile. 712 people from all Brazilian regions and biomes were interviewed between July and August 2019. Four technologies were selected by pasture specialists from Embrapa Pecuária Sudeste to be analyzed as dependent variables. The techniques were classified as adopted (1) or not adopted (0). The generalized linear model (GLM) method was used to estimate statistical parameters, and the analysis was directed to the evaluation of the combination of statistically significant independent technological processes (P &lt; 0.05). The database was then submitted to an artificial mining process, with the application of the decision tree induction method, which involves hierarchical models used for their predictive capacity. The combined use of parametric and non-parametric methodologies effectively facilitated the extraction of patterns, knowledge, and insights into stakeholders’ decision-making processes regarding the integration of technological practices in structuring pasture conservation and production systems."
},
{
"venue": "KDD",
"title": "Quantifying the contribution of triple compound extreme events to global yield loss of major staple crops from 1982 to 2016",
"authors": [
"Kun Xiao",
"Ying Sun",
"Wei‐Min Wu",
"Xuewen Zhou",
"Zhicheng Zhang",
"Qiuyao Lai",
"Chen Huang",
"Zhenhua Xiong",
"Qinchuan Xin"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.jia.2025.04.038",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.jia.2025.04.038",
"abstract": "• Assessed global crop yield responses to individual and compound extremes using a linear mixed-effects model from 1982 to 2016. • Compound events (HWLP, HDW) caused significantly greater yield losses than individual extremes, especially during critical crop growth stages. • Identified crop- and stage-specific vulnerabilities to compound extremes, emphasizing their role in global food security risk assessments. The increasing frequency of compound extreme events under ongoing climate change threatens global food security. Compared to individual extreme events, the simultaneous occurrence of multiple extreme events can exacerbate crop yield reductions, yet comprehensive assessments of these compound effects remain limited. To bridge this gap, we applied a linear mixed-effects model to quantify the impacts of individual extreme events (cold days (CD) and killing degree days (KDD)) and triple compound extreme events (heatwave and low precipitation (HWLP) and hot-dry-windy (HDW)) on the global yields of winter wheat, soybeans, and maize from 1982 to 2016. Our analysis indicated that regions severely impacted by extreme events (exceeding the 95% threshold) experienced total crop yield losses of more than 9.16, 24.89, 26.69, and 7.12% due to CD, KDD, HWLP, and HDW, respectively. The adverse effects of compound events were particularly pronounced during critical growth stages. HWLP results in yield losses of 9.4% for winter wheat and 6.8% for maize per 10 hours of exposure during the heading to harvesting stages, while soybean yields declined by 8.8% per 10 hours during the planting to three-true-leaf stage. Similarly, KDD caused a 7.4% yield reduction in winter wheat per 10°C day during the heading to harvesting stages, a 9.5% reduction in maize per 10°C day during the planting to jointing stages, and a 3.8% reduction in soybean per 10°C day during the planting to three-true-leaf stages. These findings underscore the substantial contribution of compound extreme events, which are often overlooked in existing risk assessments, in determining the global yields of major staple crops."
},
{
"venue": "KDD",
"title": "An Improved Binary Simulated Annealing Algorithm and TPE-FL-LightGBM for Fast Network Intrusion Detection",
"authors": [
"Yafei Luo",
"Ruihan Chen",
"Chuantao Li",
"Derong Yang",
"Kun Tang",
"Jing Su"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2079-9292/14/2/231/pdf?version=1736324157",
"source": "openalex",
"doi": "https://doi.org/10.3390/electronics14020231",
"abstract": "With the rapid proliferation of the Internet, network security issues that threaten users have become increasingly severe, despite the widespread benefits of Internet access. Most existing intrusion detection systems (IDS) suffer from suboptimal performance due to data imbalance and feature redundancy, while also facing high computational complexity in areas such as feature selection and optimization. To address these challenges, this study proposes a novel network intrusion detection method based on an improved binary simulated annealing algorithm (IBSA) and TPE-FL-LightGBM. First, by integrating Focal Loss into the loss function of the LightGBM classifier, we introduce cost-sensitive learning, which effectively mitigates the impact of class imbalance on model performance and enhances the model’s ability to learn difficult-to-classify samples. Next, significant improvements are made to the simulated annealing algorithm, including adaptive adjustments of the initial temperature and Metropolis criterion, the incorporation of multi-neighborhood search strategies, and the integration of an S-shaped transfer function. These improvements enable the IBSA method to achieve efficient optimal feature selection with fewer iterations. Finally, the Tree-structured Parzen Estimator (TPE) algorithm is employed to optimize the structure of the FL-LightGBM classifier, further enhancing its performance. Through comprehensive visual analysis, ablation studies, and comparative experiments on the NSL-KDD and UNSW-NB15 datasets, the reliability of the proposed network intrusion detection method is validated."
},
{
"venue": "KDD",
"title": "An Effective Method for Detecting Cyber Attacks on Computer Networks from the NSL-KDD Data Set",
"authors": [
"Aseena Babu Shaik",
"Rajeswara Reddy Saddala",
"Nagagopi Raju Vullam",
"Gondi Konda Reddy",
"Subhani Shaik"
],
"year": 2025,
"pdf_url": "https://www.itm-conferences.org/articles/itmconf/pdf/2025/05/itmconf_iccp-ci2024_02001.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1051/itmconf/20257402001",
"abstract": "Cybercrime is rapidly increasing and exploits various vulnerabilities in these computing environments. Ethical hackers pay more attention to determining vulnerabilities and recommending mitigation methods. Due to the effectiveness of machine learning in solving cybersecurity problems, machine learning is of great importance to cybersecurity. Machine learning models are used to advance the techniques to detect and solve cybersecurity problems. Machine learning methods help detect more cyber attacks more efficiently than other software-oriented techniques, reducing the burden on security analysts. Adaptive methods such as machine learning can improve detection rates. Logistic regression is used to resolve the issue of intrusion identification and a novel research model for intrusion identification. Logistic regression models can fully favor network traffic structure information to capture features more comprehensively. Experimental outcomes show that the algorithm behaves better than traditional methods."
},
{
"venue": "KDD",
"title": "An effective hybrid deep learning metaheuristic model for robust IoT intrusion detection",
"authors": [
"Alok Kumar Shukla",
"Shubhra Dwivedi",
"Aishwarya Mishra"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s10791-025-09708-w.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s10791-025-09708-w",
"abstract": "A growing body of current research highlights intrusion detection as a pivotal area of study within Internet of Things (IoT) networks, demonstrating its potential to enhance operational efficiency. The high diversity of IoT devices, their protocols and standards, and their limited computational resources have led to the appearance of novel security challenges. However, traditional security solutions to intrusion detection systems (IDS) are often unable to counter evolving new attack patterns on an overwhelming volume of records with insufficient feature diversity. The advent of increased computational power, coupled with accelerated network data traffic and reduced computing costs, has enabled researchers across disciplines to use deep learning (DL) techniques in IDS datasets to provide high accuracy and real-time detection. In this context, first, this study applies Kernel Principal Component Analysis (KPCA) to reduce the dimensionality of the dataset while preserving meaningful class distinctions. Subsequently, the Lévy flight mechanism is employed to identify functional feature sets to increase diversity in the search process. In addition, this work integrates deep learning models and addresses prevailing gaps in existing intrusion detection research, focusing on their classification performance and robustness against vulnerabilities. Additionally, we optimize Deep Neural Networks with Long Short-Term Memory architecture (DNN-LSTM) using the adaptive Lévy flight Grasshopper Optimization Algorithm (GOA) to fine-tune hyperparameters, improving anomaly detection efficacy. The proposed approach is evaluated using the CIC-IDS 2017, TON-IoT, and NSL-KDD datasets, with experimental results that indicate that the DNN-LSTM model significantly improves detection accuracy, particularly for novel or small-sample attack scenarios, outperforming standalone DL models. Empirical analysis further reveals that our approach surpasses state-of-the-art deep neural networks across several key performance metrics, underscoring its effectiveness in advancing intrusion detection research."
},
{
"venue": "KDD",
"title": "A novel hyperparameter tuning method for enhanced intrusion detection in network security",
"authors": [
"Vahid Sinap"
],
"year": 2025,
"pdf_url": "https://dergipark.org.tr/en/download/article-file/4540567",
"source": "openalex",
"doi": "https://doi.org/10.31127/tuje.1624366",
"abstract": "Intrusion Detection Systems (IDS) are essential for ensuring the security of enterprise networks and cloud-based systems, as they defend against sophisticated and evolving cyberattacks. Machine learning (ML) techniques have emerged as effective tools to enhance IDS performance, addressing the limitations of traditional methods. This study proposes a novel hyperparameter tuning method for ML-based IDS, leveraging the NSL-KDD dataset with extensive feature selection and preprocessing to address data imbalance and redundancy. The method, integrating adaptive refinement with stochastic perturbation, optimizes classifiers such as Random Forest (RF), Gradient Boosting (GB), and Extreme Gradient Boosting (XGB), achieving both higher detection accuracy (99.90% with RF) and improved computational efficiency. This approach excels due to its dynamic adjustment of parameter ranges and controlled randomness, converging faster than traditional Grid Search and Random Search by reducing iterations by up to 87.5%. The experimental results demonstrate that tree-based models, particularly RF, outperform others due to their ability to model complex, non-linear patterns, enhanced by the proposed tuning method. Measured in terms of convergence speed, CPU time, and memory usage, this method proves suitable for deployment in real-time, resource-constrained environments, offering a scalable and efficient solution for network security."
},
{
"venue": "KDD",
"title": "A Swarm-Based Multi-Objective Framework for Lightweight and Real-Time IoT Intrusion Detection",
"authors": [
"Hessah A. Alsalamah",
"Walaa N. Ismail"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2227-7390/13/15/2522/pdf?version=1754464458",
"source": "openalex",
"doi": "https://doi.org/10.3390/math13152522",
"abstract": "Internet of Things (IoT) applications and services have transformed the way people interact with their environment, enhancing comfort and quality of life. Additionally, Machine Learning (ML) approaches show significant promise for detecting intrusions in IoT environments. However, the high dimensionality, class imbalance, and complexity of network traffic—combined with the dynamic nature of sensor networks—pose substantial challenges to the development of efficient and effective detection algorithms. In this study, a multi-objective metaheuristic optimization approach, referred to as MOOIDS-IoT, is integrated with ML techniques to develop an intelligent cybersecurity system for IoT environments. MOOIDS-IoT combines a Genetic Algorithm (GA)-based feature selection technique with a multi-objective Particle Swarm Optimization (PSO) algorithm. PSO optimizes convergence speed, model complexity, and classification accuracy by dynamically adjusting the weights and thresholds of the deployed classifiers. Furthermore, PSO integrates Pareto-based multi-objective optimization directly into the particle swarm framework, extending conventional swarm intelligence while preserving a diverse set of non-dominated solutions. In addition, the GA reduces training time and eliminates redundancy by identifying the most significant input characteristics. The MOOIDS-IoT framework is evaluated using two lightweight models—MOO-PSO-XGBoost and MOO-PSO-RF—across two benchmark datasets, namely the NSL-KDD and CICIoT2023 datasets. On CICIoT2023, MOO-PSO-RF obtains 91.42% accuracy, whereas MOO-PSO-XGBoost obtains 98.38% accuracy. In addition, both models perform well on NSL-KDD (MOO-PSO-RF: 99.66% accuracy, MOO-PSO-XGBoost: 98.46% accuracy). The proposed approach is particularly appropriate for IoT applications with limited resources, where scalability and model efficiency are crucial considerations."
},
{
"venue": "KDD",
"title": "Mitigating Class Imbalance in Network Intrusion Detection with Feature-Regularized GANs",
"authors": [
"Jing Li",
"Wei Zong",
"Yang-Wai Chow",
"Willy Susilo"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/1999-5903/17/5/216/pdf?version=1747142073",
"source": "openalex",
"doi": "https://doi.org/10.3390/fi17050216",
"abstract": "Network Intrusion Detection Systems (NIDS) often suffer from severe class imbalance, where minority attack types are underrepresented, leading to degraded detection performance. To address this challenge, we propose a novel augmentation framework that integrates Soft Nearest Neighbor Loss (SNNL) into Generative Adversarial Networks (GANs), including WGAN, CWGAN, and WGAN-GP. Unlike traditional oversampling methods (e.g., SMOTE, ADASYN), our approach improves feature-space alignment between real and synthetic samples, enhancing classifier generalization on rare classes. Experiments on NSL-KDD, CSE-CIC-IDS2017, and CSE-CIC-IDS2018 show that SNNL-augmented GANs consistently improve minority-class F1-scores without degrading overall accuracy or majority-class performance. UMAP visualizations confirm that SNNL produces more compact and class-consistent sample distributions. We also evaluate the computational overhead, finding the added cost moderate. These results demonstrate the effectiveness and practicality of SNNL as a general enhancement for GAN-based data augmentation in imbalanced NIDS tasks."
},
{
"venue": "KDD",
"title": "Hybrid Cross-Temporal Contrastive Model with Spiking Energy-Efficient Network Intrusion Detection in IOMT",
"authors": [
"Fatma S. Alrayes",
"Mohammed Zakariah",
"Syed Umar Amin",
"Zafar Iqbal Khan",
"Maha Helal"
],
"year": 2025,
"pdf_url": "https://link.springer.com/content/pdf/10.1007/s44196-025-00995-1.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1007/s44196-025-00995-1",
"abstract": "The Internet of Medical Things (IoMT), a key application of the Internet of Things (IoT), has played a key role, especially during the COVID-19 pandemic. Real-time patient monitoring and remote diagnostics help improve medical services, but this increases the mammoth size of network traffic, which impacts the security quite a bit. However, traditional intrusion detection systems lack synchronization between accuracy and energy efficiency in resource-constrained IoMT environments. To address this issue, we present a Hybrid Cross-Temporal Contrastive Model coupled with a spiking energy-efficient network for intrusion detection. This approach uses contrastive learning to learn temporal dependencies in network traffic and Spiking Neural Networks (SNNs) for energy-efficient computations. We evaluated the model on the WUSTL-EHMS-2020 dataset, which consists of 44 features (35 of them are network flow measurements, and 8 are biometric patient features), as well as the NSL-KDD dataset to perform a comparative validation. Furthermore, the experiment results prove that our proposed model achieves 99.95% accuracy on the WUSTL-EHMS-2020 dataset with an F1 score of 99.89%, precision of 98.23%, and recall of 99.55%, outperforming conventional models. The model attained 98.2% accuracy, 97.6% precision, 98.5% F1 score, and 97.3% recall on the NSL-KDD dataset. Our approach shows that these results effectively secure IoMT networks at a low computational cost. Finally, the proposed hybrid model can achieve good performance and energy efficiency for intrusion detection in innovative healthcare systems. In future work, efforts will be made to improve the model’s generalization property in diverse IoMT environments and minimize the energy consumption of SNNs in real-time applications."
},
{
"venue": "KDD",
"title": "Enhanced Cyber Threat Detection System Leveraging Machine Learning Using Data Augmentation",
"authors": [
"Umar Iftikhar",
"Syed Abbas Ali"
],
"year": 2025,
"pdf_url": "https://thesai.org/Downloads/Volume16No2/Paper_23-Enhanced_Cyber_Threat_Detection_System.pdf",
"source": "openalex",
"doi": "https://doi.org/10.14569/ijacsa.2025.0160223",
"abstract": "In the modern era of cyber security, cyber-attacks are continuously evolving in terms of complexity and frequency. In this context, organizations need to enhance Network Intrusion Detection Systems (NIDS) for anomaly detection. Although the existing Machine Learning models are in place to cater to the situations but new challenges emerge rapidly which affects the performance and efficiency of existing models specifically the unreachability of large datasets and unorganized data. This results in degraded efficiency for the identification of complex attacks. In this paper, data augmentation has been done of NSL-KDD which is a standard dataset for Intrusion Detection Systems (IDS) specifically for IoT-based devices. The improvement in performance and efficiency of NIDS has been performed by training the augmented dataset using the K-Nearest Neighbor (KNN) ML model."
},
{
"venue": "KDD",
"title": "A Novel Framework of Anomaly-Based Network Intrusion Detection using Hybrid CNN, Bi-LSTM Deep Learning Techniques",
"authors": [
"Srinivas Akkepalli"
],
"year": 2025,
"pdf_url": "https://jisem-journal.com/index.php/journal/article/download/3015/1275",
"source": "openalex",
"doi": "https://doi.org/10.52783/jisem.v10i19s.3015",
"abstract": "A Novel Framework of Anomaly-based Network Intrusion Detection system using hybrid CNN,Bi-LSTM Deep learning techniques with the aim of anomaly detection, In recent years, deep learning (DL) has become increasingly important in the field of cyber security. Deep learning Algorithms efficient to detect vulnerabilities in network traffic.Objective are based on literature survey provides the various anomaly based techniques Such as NIDS,SIDS, researches are presented. [proposed CNN based BLISTM model]stand out, providing a solid basis for understanding the context of the investigation and verified results with slandered existing systems and studies.The methodology adopted for this research comprises [a hybrid CNN-based BLISTM model with Adam optimizer was used] and [the well-known NSL KDD data set was used to validate the proposed modelThe results obtained revealed that efficacy of the suggested CNN-Bi-LSTM IDS has been assessed for the NSL-KDD dataset. Imbalanced data fed into CNN-Bi-LSTM accuracy achieved 98 % recall 98% and precision 99 %, F1-score98 %, After balanced data and hyper parameter tuning of CNN-Bi-LSTM classifier, Exceptional accuracy was demonstrated by the binary classification results, which included a 99.12% accuracy for the NSL KDD dataset with the precision of 99.0%, recall of 99.26%, and F1-score of 98.11%.. Conclusions are a novel frame work enhanced accuracy in detection of anomalies in network traffic in the field of[Network security]. These implications could encompass list impacted are such as Medical and Banking, Ecommerce sites.This study contributes to the literature by hybrid CNN based BLISTM generates efficacious results. The relevance and value of this research are evidenced by comparison of generated results with existing literature results. In future work it would applied for various different datasets ."
},
{
"venue": "KDD",
"title": "Generating detectors from anomaly samples via negative selection for network intrusion detection",
"authors": [
"Zhiyong Li",
"Xiang Wei",
"Chunyan Li",
"Jianhong Sun"
],
"year": 2025,
"pdf_url": "https://www.nature.com/articles/s41598-025-20516-6.pdf",
"source": "openalex",
"doi": "https://doi.org/10.1038/s41598-025-20516-6",
"abstract": "Negative selection algorithms (NSAs), which simulate the human immune mechanism for network anomaly detection, represent a promising approach in this field. Traditional NSAs randomly generate candidate detectors in high-dimensional feature subspaces. However, network anomaly detection datasets such as NSL-KDD typically exhibit sample distributions concentrated in low-dimensional subspaces. This dimensional mismatch causes traditional NSAs to underperform due to insufficient mature detector generation. To address this limitation, we leverage underutilized anomaly samples from the training set as candidate detector centers. Since these samples encode critical information about feature space distribution, they enable effective generation of mature detectors within the relevant low-dimensional subspaces. Furthermore, to mitigate misclassification of anomalies within coverage holes, we implement secondary classification based on the class attributes of nearest neighbor samples. Experiments on the NSL-KDD and UNSW-NB15 datasets showed our method outperforming eight other algorithms."
},
{
"venue": "KDD",
"title": "Enhancing Intrusion Detection System Performance Using Reinforcement Learning : A Fairness-Aware Comparative Study on NSL-KDD and CICIDS2017",
"authors": [
"Yudhi Arta",
"Suzani Mohamad Samuri",
"Nesi Syafitri"
],
"year": 2025,
"pdf_url": "https://doaj.org/article/f78ab4bd05dc45d4a0520f9759311b8e",
"source": "openalex",
"doi": "",
"abstract": "Conventional Intrusion Detection Systems (IDS) often fail to generalize in dynamic network environments, facing challenges with evolving attack patterns and class imbalance. This study aims to evaluate and compare the effectiveness of three Reinforcement Learning (RL) paradigms to enhance IDS adaptability and accuracy against these challenges. This research employs a comparative experimental design, implementing Q-Learning, Deep Q-Networks (DQN), and Proximal Policy Optimization (PPO). These algorithms were systematically evaluated using the NSL-KDD and CICIDS2017 benchmark datasets to represent both legacy and modern network traffic. A fairness-aware evaluation framework was applied, prioritizing the Matthews Correlation Coefficient (MCC) as a primary metric alongside accuracy to ensure robust performance assessment against skewed class distributions. Experimental results demonstrate that PPO significantly outperforms value-based algorithms such as Q-Learning and DQN. On the high-dimensional CICIDS2017 dataset, PPO achieved the highest detection accuracy (96.3%) and MCC (0.913). Confusion matrix analyses confirmed PPO’s capability to simultaneously minimize false positives and false negatives. Conversely, Q-Learning exhibited poor generalization on complex data, while DQN showed improved performance due to deep value approximation but remained less stable than PPO. These findings imply that policy-gradient methods like PPO are superior for real-world IDS deployments where scalability, adaptability, and low error rates are critical. Theoretically, the results suggest that stochastic policy optimization handles complex, continuous state spaces more effectively than traditional value-estimation approaches. This study contributes a rigorous head-to-head comparative analysis of RL algorithms across multiple standard datasets using fairness-aware metrics. It bridges the research gap found in previous studies that often evaluated algorithms in isolation or relied on accuracy metrics that can be misleading in imbalanced security contexts."
},
{
"venue": "KDD",
"title": "Collaborative Intelligence for Securing Next-Generation Healthcare Systems Against Cyber Risks",
"authors": [
"Giorgia Pavani",
"K Bhaskar",
"G.V.T. Swapna",
"G Viswanath"
],
"year": 2025,
"pdf_url": "https://supublication.com/index.php/ijhsp/article/download/1995/1245/4103",
"source": "openalex",
"doi": "https://doi.org/10.47992/ijhsp.2581.6411.0133",
"abstract": "With the rapid integration of modern technologies and biotechnologies, next-generation healthcare environments are becoming increasingly dependent on interconnected smart devices. The Industry 5.0 healthcare paradigm focuses on hyper-personalization, aiming to provide human-centric, adaptive healthcare solutions through the fusion of the Internet of Things (IoT), the Internet of Medical Things (IoMT), and Artificial Intelligence (AI). This advanced paradigm allows tailored medical care for patients with diverse health conditions, improving diagnostic accuracy, treatment efficiency, and overall patient outcomes. However, with this shift toward intelligent, data-driven infrastructure comes a significant rise in cybersecurity concerns, particularly the growing vulnerability to sophisticated cyber threats targeting healthcare systems. To address these challenges, a collaborative intelligence-based intrusion detection approach has been proposed, leveraging ensemble learning techniques for real-time detection and prevention of cyber-attacks. The method utilizes the NSL-KDD dataset, a benchmark dataset for evaluating intrusion detection systems, to validate performance across multiple classifiers. The technique evaluates key machine learning algorithms, including k-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Decision Tree, and introduces a robust Stacking Classifier that integrates the strengths of Random Forest and Light Gradient Boosting Machine (LightGBM). These algorithms are assessed based on critical performance metrics such as accuracy, precision, recall, and F1-score. Experimental results reveal that the ensemble-based Stacking Classifier achieves 100% accuracy, outperforming individual classifiers and showcasing the potential of combined models in detecting anomalous network behavior effectively. This demonstrates the importance of collaborative intelligence in forming a resilient cybersecurity layer for smart healthcare applications. Such a security mechanism is vital for safeguarding sensitive medical data and maintaining trust in intelligent, automated, and highly personalized healthcare delivery systems in the Industry 5.0 era."
},
{
"venue": "KDD",
"title": "Deep Transfer Learning Based on Hybrid Swin Transformers With LSTM for Intrusion Detection Systems in IoT Environment",
"authors": [
"Ibrahim A. Fares",
"Ahmed Gamal Abdellatif",
"Mohamed Abd Elaziz",
"Mansour Shrahili",
"Adham Ahmed Elmahallawy",
"Rana Muhammad Sohaib",
"Mahmoud A. Shawky",
"Syed Tariq Shah"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1109/ojcoms.2025.3569301",
"source": "openalex",
"doi": "https://doi.org/10.1109/ojcoms.2025.3569301",
"abstract": "Extensive growth in the number of Internet Of Things (IoT) devices has significantly increased susceptibility to various cyber-attacks and hence emphasized the need for robust intrusion detection systems (IDS) for ensuring IoT network security. While deep learning (DL) methodologies have proven effective in the application of IDS, their success greatly depends on the availability of large datasets and significant computational resources during training. To overcome the limitations associated with this dependence on large datasets and significant computational capacity for training, the current work suggests employing the transfer learning (TL) mechanism by combining Swin Transformers with long short-term memory (LSTM) networks. Utilizing the beneficial properties of Swin Transformers in learning hierarchically structured data combined with the proficiency of LSTM in processing sequential dependencies, the hybrid model generates pre-trained weights in the first phase. These pre-trained weights are further transferred into another instance of the new model for subsequent fine-tuning. Experiments are carried out on several benchmarking datasets, namely NSL-KDD, ToN-IoT, BoTIoT, MQTTIoT, and CICIoT2023, which include both binary and multi-class classification scenarios. The proposed model outperforms state-of-the-art DL models, for example, the Autoencoders, ResNets, CNN, RNN, and LSTM models, and achieved an average of 98.97% in accuracy, of 98.97% in precision, of 99.02% in recall, of 98.97% in F1 score, across all datasets. Experimental results establish that the hybrid approach achieves better detection accuracy and better performance measures compared to the latest state-of-the-art methods, thus proving itself effective in increasing the scalability and adaptability of IDS in IoT."
},
{
"venue": "KDD",
"title": "Quantum-Driven Chaos-Informed Deep Learning Framework for Efficient Feature Selection and Intrusion Detection in IoT Networks",
"authors": [
"Padmasri Turaka",
"Saroj Kumar Panigrahy"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2227-7080/13/10/470/pdf?version=1760710504",
"source": "openalex",
"doi": "https://doi.org/10.3390/technologies13100470",
"abstract": "The rapid development of the Internet of Things (IoT) poses significant problems in securing heterogeneous, massive, and high-volume network traffic against cyber threats. Traditional intrusion detection systems (IDSs) are often found to be poorly scalable, or are ineffective computationally, because of the presence of redundant or irrelevant features, and they suffer from high false positive rates. Addressing these limitations, this study proposes a hybrid intelligent model that combines quantum computing, chaos theory, and deep learning to achieve efficient feature selection and effective intrusion classification. The proposed system offers four novel modules for feature optimization: chaotic swarm intelligence, quantum diffusion modeling, transformer-guided ranking, and multi-agent reinforcement learning, all of which work with a graph-based classifier enhanced with quantum attention mechanisms. This architecture allows as much as 75% feature reduction, while achieving 4% better classification accuracy and reducing computational overhead by 40% compared to the best-performing models. When evaluated on benchmark datasets (NSL-KDD, CICIDS2017, and UNSW-NB15), it shows superior performance in intrusion detection tasks, thereby marking it as a viable candidate for scalable and real-time IoT security analytics."
},
{
"venue": "KDD",
"title": "Securing Internet of Things Devices with Federated Learning: A Privacy-Preserving Approach for Distributed Intrusion Detection",
"authors": [
"Sulaiman Al Amro"
],
"year": 2025,
"pdf_url": "https://doi.org/10.32604/cmc.2025.063734",
"source": "openalex",
"doi": "https://doi.org/10.32604/cmc.2025.063734",
"abstract": "The rapid proliferation of Internet of Things (IoT) devices has heightened security concerns, making intrusion detection a pivotal challenge in safeguarding these networks. Traditional centralized Intrusion Detection Systems (IDS) often fail to meet the privacy requirements and scalability demands of large-scale IoT ecosystems. To address these challenges, we propose an innovative privacy-preserving approach leveraging Federated Learning (FL) for distributed intrusion detection. Our model eliminates the need for aggregating sensitive data on a central server by training locally on IoT devices and sharing only encrypted model updates, ensuring enhanced privacy and scalability without compromising detection accuracy. Key innovations of this research include the integration of advanced deep learning techniques for real-time threat detection with minimal latency and a novel model to fortify the system’s resilience against diverse cyber-attacks such as Distributed Denial of Service (DDoS) and malware injections. Our evaluation on three benchmark IoT datasets demonstrates significant improvements: achieving 92.78% accuracy on NSL-KDD, 91.47% on BoT-IoT, and 92.05% on UNSW-NB15. The precision, recall, and F1-scores for all datasets consistently exceed 91%. Furthermore, the communication overhead was reduced to 85 MB for NSL-KDD, 105 MB for BoT-IoT, and 95 MB for UNSW-NB15—substantially lower than traditional centralized IDS approaches. This study contributes to the domain by presenting a scalable, secure, and privacy-preserving solution tailored to the unique characteristics of IoT environments. The proposed framework is adaptable to dynamic and heterogeneous settings, with potential applications extending to other privacy-sensitive domains. Future work will focus on enhancing the system’s efficiency and addressing emerging challenges such as model poisoning attacks in federated environments."
},
{
"venue": "KDD",
"title": "A Deep Learning-Based Framework for Real-Time Detection of Cybersecurity Threats in IoT Environments",
"authors": [
"Sultan Ahmed Almalki"
],
"year": 2025,
"pdf_url": "http://thesai.org/Downloads/Volume16No3/Paper_43-A_Deep_Learning_Based_Framework_for_Real_Time_Detection.pdf",
"source": "openalex",
"doi": "https://doi.org/10.14569/ijacsa.2025.0160343",
"abstract": "The rapid adoption of Internet of Things (IoT) devices has led to an exponential increase in cybersecurity threats, necessitating efficient and real-time intrusion detection systems (IDS). Traditional IDS and machine learning models struggle with evolving attack patterns, high false positive rates, and computational inefficiencies in IoT environments. This study proposes a deep learning-based framework for real-time detection of cybersecurity threats in IoT networks, leveraging Transformers, Convolutional Neural Networks (CNNs), and Long Short-Term Memory (LSTM) architectures. The proposed framework integrates hybrid feature extraction techniques, enabling accurate anomaly detection while ensuring low latency and high scalability for IoT devices. Experimental evaluations on benchmark IoT security datasets (CICIDS2017, NSL-KDD, and TON_IoT) demonstrate that the Transformer-based model outperforms conventional IDS solutions, achieving 98.3% accuracy with a false positive rate as low as 1.9%. The framework also incorporates adversarial defense mechanisms to enhance resilience against evasion attacks. The results validate the efficacy, adaptability, and real-time applicability of the proposed deep learning approach in securing IoT networks against cyber threats."
},
{
"venue": "KDD",
"title": "IMPLEMENTASI K-MEANS CLUSTERING DALAM PENGELOMPOKAN DATA KUNJUNGAN WISATAWAN ASING DI INDONESIA",
"authors": [
"Miftahul Arif Aldi",
"Zaehol Fatah"
],
"year": 2025,
"pdf_url": "https://journal.smartpublisher.id/index.php/jimi/article/download/336/342",
"source": "openalex",
"doi": "https://doi.org/10.69714/3hhfj353",
"abstract": "Clustering is a data mining technique used for grouping data based on specific similarities. This study implements K-Means Clustering to analyze foreign tourist visit data in Indonesia in 2024. Using the Knowledge Discovery in Database (KDD) methodology, the research involves five stages: Data Selection, preprocessing, Transformation, data mining, and Evaluation. Data Clustering was conducted using RapidMiner software, experimenting with different cluster counts (k=2 to k=7) to determine the optimal number of clusters. Results indicate that three clusters (k=3) with the smallest Davies-Bouldin Index (DBI) value were optimal. This Clustering approach categorizes tourists into low, medium, and high visit groups, assisting policymakers in strategic tourism development. The findings support capacity planning and seasonal marketing strategies to optimize Indonesia's tourism sector."
},
{
"venue": "KDD",
"title": "Deep Reinforcement Learning-based Asymmetric Convolutional Autoencoder for Intrusion Detection",
"authors": [
"Yuxing Dai",
"Xinjie Qian",
"Chunmei Yang"
],
"year": 2025,
"pdf_url": "https://journals.riverpublishers.com/index.php/JICTS/article/download/28429/21915",
"source": "openalex",
"doi": "https://doi.org/10.13052/jicts2245-800x.1314",
"abstract": "In recent years, intrusion detection systems (IDSs) have become a critical component of network security, due to the growing number and complexity of cyber-attacks. Traditional IDS methods, including signature-based and anomaly-based detection, often struggle with the high-dimensional and imbalanced nature of network traffic, leading to suboptimal performance. Moreover, many existing models fail to efficiently handle the diverse and complex attack types. In response to these challenges, we propose a novel deep learning-based IDS framework that leverages a deep asymmetric convolutional autoencoder (DACA) architecture. Our model combines advanced techniques for feature extraction, dimensionality reduction, and anomaly detection into a single cohesive framework. The DACA model is designed to effectively capture complex patterns and subtle anomalies in network traffic while significantly reducing computational complexity. By employing this architecture, we achieve superior detection accuracy across various types of attacks even in imbalanced datasets. Experimental results demonstrate that our approach surpasses several state-of-the-art methods, including HCM-SVM, D1-IDDS, and GNN -IDS, achieving high accuracy, precision, recall, and F1-score on benchmark datasets such as NSL-KDD and UNSW-NB15. The results emphasize how effectively our model identifies complex and varied attack patterns. In conclusion, the proposed IDS model offers a promising solution to the limitations of current detection systems, with significant improvements in performance and efficiency. This approach contributes to advancing the development of robust and scalable network security solutions."
},
{
"venue": "KDD",
"title": "Adaptive Network Intrusion Detection Using Reinforcement Learning with Proximal Policy Optimization",
"authors": [
"Akshaya Suresh",
"Arun Cyril Jose"
],
"year": 2025,
"pdf_url": "https://dl.acm.org/doi/pdf/10.1145/3764586",
"source": "openalex",
"doi": "https://doi.org/10.1145/3764586",
"abstract": "In an increasingly digital and interconnected world, the need for robust network intrusion detection systems is crucial to ensure cybersecurity. This article presents a novel approach to network intrusion detection that integrates both traditional machine learning methods and advanced reinforcement learning techniques to enhance detection capabilities and accuracy. The proposed system uses Proximal Policy Optimization, a reinforcement learning algorithm, to dynamically adjust ensemble weights, thereby optimizing the contributions of base learners, such as Random Forest and CatBoost. Additionally, a Multi-layer Perceptron-based meta-learner is employed to refine the predictions, leading to an overall improvement in detection performance. The model was evaluated on five diverse datasets, including NSL-KDD, CICIDS, TON IoT, DDoS, and UNSW-NB15, achieving an average accuracy of 97.16%, and an average precision, recall, and F1-score of 97% across all datasets. The proposed work is compared with the existing state-of-the-art detection methods demonstrating its better performance in detecting both known and novel attack types. Furthermore, the integration of reinforcement learning allowed for dynamic and context-sensitive decision-making, enabling the system to handle complex attack patterns that traditional models struggle with. The training and validation results across all datasets showed rapid convergence and minimal overfitting, further supporting the model’s robustness."
},
{
"venue": "KDD",
"title": "MIDS-GAN: Minority Intrusion Data Synthesizer GAN—An ACON Activated Conditional GAN for Minority Intrusion Detection",
"authors": [
"Chalerm Klinkhamhom",
"Pongsarun Boonyopakorn",
"Pongpisit Wuttidittachotti"
],
"year": 2025,
"pdf_url": "https://doi.org/10.3390/math13213391",
"source": "openalex",
"doi": "https://doi.org/10.3390/math13213391",
"abstract": "Intrusion Detection Systems (IDS) are vital to cybersecurity but suffer from severe class imbalance in benchmark datasets such as NSL-KDD and UNSW-NB15. Conventional oversampling methods (e.g., SMOTE, ADASYN) are efficient yet fail to preserve the latent semantics of rare attack behaviors. This study introduces the Minority-class Intrusion Detection Synthesizer GAN (MIDS-GAN), a divergence-minimization framework for minority data augmentation under structured feature constraints. MIDS-GAN integrates (i) correlation-based structured feature selection (SFS) to reduce redundancy, (ii) trainable ACON activations to enhance generator expressiveness, and (iii) KL-divergence-guided alignment to ensure distributional fidelity. Experiments on NSL-KDD and UNSW-NB15 demonstrate significant improvement on detection, with recall increasing from 2% to 27% for R2L and 1% to 17% for U2R in NSL-KDD, and from 18% to 44% for Worms and 69% to 75% for Shellcode in UNSW-NB15. Weighted F1-scores also improved to 78%, highlighting MIDS-GAN’s effectiveness in enhancing minority-class detection through a principled, divergence-aware approach."
},
{
"venue": "KDD",
"title": "Provenance Graph-Based Deep Learning Framework for APT Detection in Edge Computing",
"authors": [
"Tianyi Wang",
"Wei Tang",
"Yuan Su",
"Jiliang Li"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/2076-3417/15/16/8833/pdf?version=1754898827",
"source": "openalex",
"doi": "https://doi.org/10.3390/app15168833",
"abstract": "Edge computing builds relevant services and applications on the edge server near the user side, which enables a faster service response. However, the lack of large-scale hardware resources leads to weak defense for edge devices. Therefore, proactive defense security mechanisms, such as Intrusion Detection Systems (IDSs), are widely deployed in edge computing. Unfortunately, most of those IDSs lack causal analysis capabilities and still suffer the threats from Advanced Persistent Threat (APT) attacks. To effectively detect APT attacks, we propose a heterogeneous graph neural networks threat detection model based on the provenance graph. Specifically, we leverage the powerful analysis and tracking capabilities of the provenance graph to model the long-term behavior of the adversary. Moreover, we leverage the predictive power of heterogeneous graph neural networks to embed the provenance graph by a node-level and semantic-level heterogeneous mutual attention mechanism. In addition, we also propose a provenance graph reduction algorithm based on the semantic similarity of graph substructures to improve the detection efficiency and accuracy of the model, which reduces and integrates redundant information by calculating the semantic similarity between substructures. The experimental results demonstrate that the prediction accuracy of our method reaches 99.8% on the StreamSpot dataset and achieves 98.13% accuracy on the NSL-KDD dataset."
},
{
"venue": "KDD",
"title": "Attention-driven multi-model architecture for unbalanced network traffic intrusion detection via extreme gradient boosting",
"authors": [
"Oluwadamilare Harazeem Abdulganiyu",
"Taha Ait Tchakoucht",
"A. El Alaoui",
"Yakub Kayode Saheed"
],
"year": 2025,
"pdf_url": "https://doi.org/10.1016/j.iswa.2025.200519",
"source": "openalex",
"doi": "https://doi.org/10.1016/j.iswa.2025.200519",
"abstract": "• In order to address the security challenge inherent in the context of intrusion detection in imbalanced network traffic, this study proposed an Attention-Driven Multi-Model Architecture called CWFLAM-VAE integrated with Extreme Gradient Boosting (XGBoost) for Detecting Intrusions in Imbalance Network Traffic. While both techniques have been used individually in the context of intrusion detection, their combined application in this manner has not been previously explored which makes it a novel approach to tackling the problem of class imbalance in Network Intrusion Detection Systems (NIDS). • We designed a class-wise focal loss attention mechanism-based variational auto-encoder (CWFLAM-VAE) to tackle the issue of class imbalance. The attention mechanism makes it possible for the model to generate each element of the output sequence while concentrating on specific areas of the input sequence. The class-wise focal loss was designed to augment the underrepresented minority samples by allocating different weights to different classes, with a more specific focus on the rare samples. • We employed Extreme Gradient Boosting (XGBoost) to train the proposed framework's key features by enhancing the efficacy of the IDS by improving detection, while also diminishing the duration of both the training and overall detections time. • The effectiveness of the proposed IDS model was compared with baseline classifiers and data balancing techniques, as well as with the relevant state-of-the-art work, which depict that the proposed model offers better performance in the overall detection of intrusions of minority samples in an imbalanced network traffic. Network Intrusion Detection Systems (NIDS) face significant challenges in identifying rare attack instances due to the inherent class imbalance and diversity in network traffic. This imbalance, often characterized by a dominance of benign network traffic data, reduces the effectiveness of traditional detection methods. To address this, we proposed CWFLAM-VAE, an attention-driven multi-model architecture that combines Class-Wise Focal Loss, Variational Autoencoder, and Extreme Gradient Boosting. CWFLAM-VAE generates synthetic rare-class attack data while preserving the original feature distribution, mitigating imbalance and improving classification performance. The effectiveness of our proposed system was evaluated by employing two datasets, one of which is the NSL-KDD, which exhibits a skewed distribution of network traffic favoring the majority class, and CSE-CIC-IDS2018 dataset, where approximately 83 % of the data consists of benign network traffic. We compared our method with existing sampling techniques (SMOTE, ROS, ADASYN, RUS) and existing classifiers (Logistic Regression, KNN, SVM, Decision Tree, LSTM, CNN). The experimental findings distinctly reveal the efficacy of the CWFLAM-VAE in resolving class imbalance concerns, with Extreme Gradient Boosting surpassing alternative machine learning techniques particularly in the detection of rare instances of attack traffic with an f-score of 97.6 % and 98.1 %, as well as a false positive rate of 0.17 and 0.27 for both data respectively."
},
{
"venue": "KDD",
"title": "Enhanced intrusion detection in smart grids using extended long short-term memory variants",
"authors": [
"Saida Baalia",
"Djalila Boughareb",
"Zineddine Kouahla",
"Hamid Seridi"
],
"year": 2025,
"pdf_url": "https://ijain.org/index.php/IJAIN/article/download/2169/ijain_vol11no4pp734-755",
"source": "openalex",
"doi": "https://doi.org/10.26555/ijain.v11i4.2169",
"abstract": "Smart grid systems, which integrate traditional energy infrastructure with modern communication technologies, face significant cybersecurity challenges due to their dynamic architecture and continuous data exchange. The diversity and interconnection of devices increase vulnerability to malicious intrusions, highlighting the need for advanced and scalable detection methods. This study aims to develop an intrusion detection system (IDS) for smart grids by leveraging recent advances in deep learning, specifically enhanced variants of Long Short-Term Memory (LSTM)—xLSTM, sLSTM, and mLSTM. These sequence modeling architectures were adapted and fine-tuned within our IDS framework to capture complex spatio-temporal patterns and handle heterogeneous, high-dimensional data effectively. A comprehensive evaluation on two benchmark datasets, NSL-KDD and DNP3, demonstrates the robustness of the proposed approach. On the NSL- KDD, xLSTM, sLSTM, and mLSTM achieved accuracies of 98.16%, 98.55%, and 98.54%. On the more modern, protocol-specific DNP3 dataset, which represents real-world SCADA-focused attacks, the models maintained their superior performance, achieving accuracies of 99.50%, 99.33%, and 99.42%, respectively. The high and consistent accuracy across both datasets demonstrates the models' dependability and adaptability for intrusion detection in smart grid infrastructures. The study's targeted enhancement of LSTM-based architectures contributes a novel and effective approach to protecting critical intelligent systems from emerging cyber threats."
},
{
"venue": "KDD",
"title": "IGSA-SAC: a novel approach for intrusion detection using improved gravitational search algorithm and soft actor-critic",
"authors": [
"Lizhong Jin",
"Rulong Fan",
"Xiaoling Han",
"Xueying Cui"
],
"year": 2025,
"pdf_url": "https://doi.org/10.3389/fcomp.2025.1574211",
"source": "openalex",
"doi": "https://doi.org/10.3389/fcomp.2025.1574211",
"abstract": "Background Network intrusion detection is a critical component of maintaining network security, especially as cyber threats become increasingly sophisticated. While deep learning-based intrusion detection algorithms have shown promise, they often struggle with high-dimensional datasets containing outliers, anomalies, or rare events. This study addresses these challenges by proposing a novel approach that combines the Improved Gravitational Search Algorithm (IGSA) with the Soft Actor-Critic (SAC) reinforcement learning algorithm, aiming to enhance detection accuracy and computational efficiency. Methods We introduce the IGSA-SAC intrusion detection model, which leverages an enhanced Gravitational Search Algorithm (IGSA) to improve robustness against outliers and dynamically adjust the exploration-exploitation balance. This is achieved through fitness normalization with an Adaptive Search Radius and a sigmoid function to modulate the gravitational constant. The IGSA-SAC method effectively navigates the search space to identify the most relevant features for intrusion detection, reducing dimensionality and computational complexity. Additionally, we design a reinforcement learning reward function to guide the learning process, encouraging the agent to improve detection effectiveness while minimizing false alarms and missed detections. Results Experiments were conducted on the NSL-KDD and AWID datasets to evaluate the performance of IGSA-SAC. The results demonstrate that IGSA-SAC achieves an accuracy of 84.15% and an F 1-score of 84.85% on the NSL-KDD dataset. On the AWID dataset, IGSA-SAC surpasses 98.9% in both accuracy and F 1-score, outperforming existing intrusion detection algorithms. Conclusions The proposed IGSA-SAC method significantly improves intrusion detection performance by effectively handling high-dimensional datasets and reducing computational complexity. The results highlight the potential of IGSA-SAC as a robust and efficient solution for real-world network intrusion detection systems, offering enhanced accuracy and reliability in identifying cyber threats."
},
{
"venue": "KDD",
"title": "AS-TBR: An Intrusion Detection Model for Smart Grid Advanced Metering Infrastructure",
"authors": [
"Hao Ma",
"Yifan Fan",
"Yiying Zhang"
],
"year": 2025,
"pdf_url": "https://www.mdpi.com/1424-8220/25/10/3155/pdf?version=1747643009",
"source": "openalex",
"doi": "https://doi.org/10.3390/s25103155",
"abstract": "Advanced Metering Infrastructure (AMI), as a critical data collection and communication hub within the smart grid architecture, is highly vulnerable to network intrusions due to its open bidirectional communication network. A significant challenge in AMI traffic data is the severe class imbalance, where existing methods tend to favor majority class samples while neglecting the detection of minority class attacks, thereby undermining the overall reliability of the detection system. Additionally, current approaches exhibit limitations in spatiotemporal feature extraction, failing to effectively capture the complex dependencies within network traffic data. In terms of global dependency modeling, existing models struggle to dynamically adjust key features, impacting the efficiency and accuracy of intrusion detection and response. To address these issues, this paper proposes an innovative hybrid deep learning model, AS-TBR, for AMI intrusion detection in smart grids. The proposed model incorporates the Adaptive Synthetic Sampling (ADASYN) technique to mitigate data imbalance, thereby enhancing the detection accuracy of minority class samples. Simultaneously, Transformer is leveraged to capture global temporal dependencies, BiGRU is employed to model bidirectional temporal relationships, and ResNet is utilized for deep spatial feature extraction. Experimental results demonstrate that the AS-TBR model achieves an accuracy of 93% on the UNSW-NB15 dataset and 80% on the NSL-KDD dataset. Furthermore, it outperforms baseline models in terms of precision, recall, and other key evaluation metrics, validating its effectiveness and robustness in AMI intrusion detection."
},
{
"venue": "ICML",
"title": "KGMark: A Diffusion Watermark for Knowledge Graphs",
"authors": [
"Hongrui Peng",
"Haolang Lu",
"Yuanlong Yu",
"WeiYe Fu",
"Kun Wang",
"Guoshun Nan"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=GKZySvM2t9",
"source": "openreview",
"doi": "",
"abstract": "Knowledge graphs (KGs) are ubiquitous in numerous real-world applications, and watermarking facilitates protecting intellectual property and preventing potential harm from AI-generated content. Existing watermarking methods mainly focus on static plain text or image data, while they can hardly be applied to dynamic graphs due to spatial and temporal variations of structured data. This motivates us to propose KGMark, the first graph watermarking framework that aims to generate robust, detectable, and transparent diffusion fingerprints for dynamic KG data. Specifically, we propose a novel clustering-based alignment method to adapt the watermark to spatial variations. Meanwhile, we present a redundant embedding strategy to harden the diffusion watermark against various attacks, facilitating the robustness of the watermark to the temporal variations. Additionally, we introduce a novel learnable mask matrix to improve the transparency of diffusion fingerprints. By doing so, our KGMark properly tackles the variation challenges of structured data. Experiments on various public benchmarks show the effectiveness of our proposed KGMark."
},
{
"venue": "ICML",
"title": "Emoji Attack: Enhancing Jailbreak Attacks Against Judge LLM Detection",
"authors": [
"Zhipeng Wei",
"Yuqi Liu",
"N. Benjamin Erichson"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=Q0rKYiVEZq",
"source": "openreview",
"doi": "",
"abstract": "Jailbreaking techniques trick Large Language Models (LLMs) into producing restricted output, posing a potential threat. One line of defense is to use another LLM as a Judge to evaluate the harmfulness of generated text. However, we reveal that these Judge LLMs are vulnerable to token segmentation bias, an issue that arises when delimiters alter the tokenization process, splitting words into smaller sub-tokens. This alters the embeddings of the entire sequence, reducing detection accuracy and allowing harmful content to be misclassified as safe. In this paper, we introduce Emoji Attack, a novel strategy that amplifies existing jailbreak prompts by exploiting token segmentation bias. Our method leverages in-context learning to systematically insert emojis into text before it is evaluated by a Judge LLM, inducing embedding distortions that significantly lower the likelihood of detecting unsafe content. Unlike traditional delimiters, emojis also introduce semantic ambiguity, making them particularly effective in this attack. Through experiments on state-of-the-art Judge LLMs, we demonstrate that Emoji Attack substantially reduces the unsafe prediction rate, bypassing existing safeguards."
},
{
"venue": "ICML",
"title": "HiRemate: Hierarchical Approach for Efficient Re-materialization of Neural Networks",
"authors": [
"Julia Gusak",
"Xunyi Zhao",
"Théotime Le Hellard",
"Zhe LI",
"Lionel Eyraud-Dubois",
"Olivier Beaumont"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=rnx11J4hsg",
"source": "openreview",
"doi": "",
"abstract": "Training deep neural networks (DNNs) on memory-limited GPUs is challenging, as storing intermediate activations often exceeds available memory. Re-materialization, a technique that preserves exact computations, addresses this by selectively recomputing activations instead of storing them. However, existing methods either fail to scale, lack generality, or introduce excessive execution overhead. We introduce ${\\mbox{HiRemate}}$ a ${\\textit hierarchical}$ re-materialization framework that recursively partitions large computation graphs, applies optimized solvers at multiple levels, and merges solutions into a global efficient training schedule. This enables scalability to significantly larger graphs than prior ILP-based methods while keeping runtime overhead low. Designed for single-GPU models and activation re-materialization, HiRemate extends the feasibility of training networks with thousands of graph nodes, surpassing prior methods in both efficiency and scalability. Experiments on various types of networks yield up to 50-70% memory reduction with only 10-15% overhead, closely matching optimal solutions while significantly reducing solver time. Seamlessly integrating with PyTorch Autograd, HiRemate requires almost no code change to use, enabling broad adoption in memory-constrained deep learning."
},
{
"venue": "ICML",
"title": "ITFormer: Bridging Time Series and Natural Language for Multi-Modal QA with Large-Scale Multitask Dataset",
"authors": [
"Yilin wang",
"Peixuan Lei",
"Jie Song",
"Yuzhe Hao",
"Tao Chen",
"Yuxuan Zhang",
"LEI JIA",
"Yuanxiang Li",
"zhongyu wei"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=GByP03IitA",
"source": "openreview",
"doi": "",
"abstract": "Time-series data are critical in diverse applications, such as industrial monitoring, medical diagnostics, and climate research. However, effectively integrating these high-dimensional temporal signals with natural language for dynamic, interactive tasks remains a significant challenge. To address this, we introduce the Time-Series Question Answering (Time-Series QA) task and release EngineMT-QA, the first large-scale, multi-task, temporal-textual QA dataset designed to capture complex interactions between time-series signals and natural language. Building on this resource, we propose the Instruct Time Transformer (ITFormer), a novel framework that bridges time-series encoders with frozen large language models (LLMs). ITFormer effectively extracts, aligns, and fuses temporal and textual features, achieving a strong improvement in QA accuracy over strong baselines with fewer than 1\\% additional trainable parameters. By combining computational efficiency with robust cross-modal modeling, our work establishes a adaptable paradigm for integrating temporal data with natural language, paving the way for new research and applications in multi-modal AI. More details about the project, including datasets and code, are available at: https://pandalin98.github.io/itformer_site/."
},
{
"venue": "ICML",
"title": "GPEN: Global Position Encoding Network for Enhanced Subgraph Representation Learning",
"authors": [
"Nannan Wu",
"Yuming Huang",
"Yiming Zhao",
"Jie Chen",
"Wenjun Wang"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=7QFmZ7i7sr",
"source": "openreview",
"doi": "",
"abstract": "Subgraph representation learning has attracted growing interest due to its wide applications in various domains. However, existing methods primarily focus on local neighborhood structures while overlooking the significant impact of global structural information, in particular the influence of multi-hop neighbors beyond immediate neighborhoods. This presents two key challenges: how to effectively capture the structural relationships between distant nodes, and how to prevent excessive aggregation of global structural information from weakening the discriminative ability of subgraph representations.\nTo address these challenges, we propose GPEN (Global Position Encoding Network). GPEN leverages a hierarchical tree structure to encode each node's global position based on its path distance to the root node, enabling a systematic way to capture relationships between distant nodes. Furthermore, we introduce a boundary-aware convolution module that selectively integrates global structural information while maintaining the unique structural patterns of each subgraph. Extensive experiments on eight public datasets identify that GPEN significantly outperforms state-of-the-art methods in subgraph representation learning."
},
{
"venue": "ICML",
"title": "Pairwise Maximum Likelihood For Multi-Class Logistic Regression Model With Multiple Rare Classes",
"authors": [
"Xuetong Li",
"Danyang Huang",
"Hansheng Wang"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=9Kywz2fO26",
"source": "openreview",
"doi": "",
"abstract": "We study in this work the problem of multi-class logistic regression with one major class and multiple rare classes, which is motivated by a real application in TikTok live stream data. The model is inspired by the two-class logistic regression model of Wang (2020) but with surprising theoretical findings, which in turn motivate new estimation methods with excellent statistical and computational efficiency. \nSpecifically, since rigorous theoretical analysis suggests that the resulting maximum likelihood estimators of different rare classes should be asymptotically independent, we consider to solve multiple pairwise two-class logistic regression problems instead of optimizing the joint log-likelihood function with computational challenge in multi-class problem, which are computationally much easier and can be conducted in a fully parallel way. To further reduce the computation cost, a subsample-based pairwise likelihood estimator is developed by down-sampling the major class. We show rigorously that the resulting estimators could be as asymptotically efficient as the global maximum likelihood estimator under appropriate regularity conditions. Extensive simulation studies are presented to support our theoretical findings and a TikTok live stream dataset is analyzed for illustration purpose."
},
{
"venue": "ICML",
"title": "ShieldAgent: Shielding Agents via Verifiable Safety Policy Reasoning",
"authors": [
"Zhaorun Chen",
"Mintong Kang",
"Bo Li"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=DkRYImuQA9",
"source": "openreview",
"doi": "",
"abstract": "Autonomous agents powered by foundation models have seen widespread adoption across various real-world applications. However, they remain highly vulnerable to malicious instructions and attacks, which can result in severe consequences such as privacy breaches and financial losses. More critically, existing guardrails for LLMs are not applicable due to the complex and dynamic nature of agents. To tackle these challenges, we propose ShieldAgent, the first guardrail agent designed to enforce explicit safety policy compliance for the action trajectory of other protected agents through logical reasoning. Specifically, ShieldAgent first constructs a safety policy model by extracting verifiable rules from policy documents and structuring them into a set of action-based probabilistic rule circuits. Given the action trajectory of the protected agent, ShieldAgent retrieves relevant rule circuits and generates a shielding plan, leveraging its comprehensive tool library and executable code for formal verification. In addition, given the lack of guardrail benchmarks for agents, we introduce ShieldAgent-Bench, a dataset with 3K safety-related pairs of agent instructions and action trajectories, collected via SOTA attacks across 6 web environments and 7 risk categories. Experiments show that ShieldAgent achieves SOTA on ShieldAgent-Bench and three existing benchmarks, outperforming prior methods by 11.3% on average with a high recall of 90.1%. Additionally, ShieldAgent reduces API queries by 64.7% and inference time by 58.2%, demonstrating its high precision and efficiency in safeguarding agents. Our project is available and continuously maintained here: https://shieldagent-aiguard.github.io/"
},
{
"venue": "ICML",
"title": "Rethinking Benign Overfitting in Two-Layer Neural Networks",
"authors": [
"Ruichen Xu",
"Kexin Chen"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=Uc0dTE2Wox",
"source": "openreview",
"doi": "",
"abstract": "Recent theoretical studies (Kou et al., 2023; Cao et al., 2022) revealed a sharp phase transition from benign to harmful overfitting when the\nnoise-to-feature ratio exceeds a threshold—a situation common in long-tailed data distributions where atypical data is prevalent. However, such harmful overfitting rarely happens in overparameterized neural networks. Further experimental results suggested that memorization is necessary for achieving near-optimal generalization error in long-tailed data distributions (Feldman & Zhang, 2020). We argue that this discrepancy between theoretical predictions and empirical observations arises because previous feature-noise data models overlook the heterogeneous nature of noise across different data classes. In this paper, we refine the feature-noise data model by incorporating class-dependent heterogeneous noise and re-examine the overfitting phenomenon in neural networks. Through a comprehensive analysis of the training dynamics, we establish test loss bounds for the refined model. Our findings reveal that neural networks can leverage \"data noise\" to learn implicit features that improve the classification accuracy for long-tailed data. Our analysis also provides a training-free metric for evaluating data influence on test performance. Experimental validation on both synthetic and real-world datasets supports our theoretical results."
},
{
"venue": "ICML",
"title": "Symmetry-Aware GFlowNets",
"authors": [
"Hohyun Kim",
"Seunggeun Lee",
"Min-hwan Oh"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=JD4eHocSPi",
"source": "openreview",
"doi": "",
"abstract": "Generative Flow Networks (GFlowNets) offer a powerful framework for sampling graphs in proportion to their rewards. However, existing approaches suffer from systematic biases due to inaccuracies in state transition probability computations. These biases, rooted in the inherent symmetries of graphs, impact both atom-based and fragment-based generation schemes. To address this challenge, we introduce Symmetry-Aware GFlowNets (SA-GFN), a method that incorporates symmetry corrections into the learning process through reward scaling. By integrating bias correction directly into the reward structure, SA-GFN eliminates the need for explicit state transition computations. Empirical results show that SA-GFN enables unbiased sampling while enhancing diversity and consistently generating high-reward graphs that closely match the target distribution."
},
{
"venue": "ICML",
"title": "CombiMOTS: Combinatorial Multi-Objective Tree Search for Dual-Target Molecule Generation",
"authors": [
"Thibaud Southiratn",
"Bonil Koo",
"Yijingxiu Lu",
"Sun Kim"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=FSlTEObdLl",
"source": "openreview",
"doi": "",
"abstract": "Dual-target molecule generation, which focuses on discovering compounds capable of interacting with two target proteins, has garnered significant attention due to its potential for improving therapeutic efficiency, safety and resistance mitigation.\nExisting approaches face two critical challenges.\nFirst, by simplifying the complex dual-target optimization problem to scalarized combinations of individual objectives, they fail to capture important trade-offs between target engagement and molecular properties. \nSecond, they typically do not integrate synthetic planning into the generative process.\nThis highlights a need for more appropriate objective function design and synthesis-aware methodologies tailored to the dual-target molecule generation task.\nIn this work, we propose CombiMOTS, a Pareto Monte Carlo Tree Search (PMCTS) framework that generates dual-target molecules.\nCombiMOTS is designed to explore a synthesizable fragment space while employing vectorized optimization constraints to encapsulate target affinity and physicochemical properties.\nExtensive experiments on real-world databases demonstrate that CombiMOTS produces novel dual-target molecules with high docking scores, enhanced diversity, and balanced pharmacological characteristics, showcasing its potential as a powerful tool for dual-target drug discovery.\nThe code and data is accessible through \\url{https://github.com/Tibogoss/CombiMOTS}."
},
{
"venue": "ICML",
"title": "Scalable Meta-Learning via Mixed-Mode Differentiation",
"authors": [
"Iurii Kemaev",
"Dan A. Calian",
"Luisa M Zintgraf",
"Gregory Farquhar",
"Hado van Hasselt"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=NWKjVzkDzg",
"source": "openreview",
"doi": "",
"abstract": "Gradient-based bilevel optimisation is a powerful technique with applications in hyperparameter optimisation, task adaptation, algorithm discovery, meta-learning more broadly, and beyond. It often requires differentiating through the gradient-based optimisation process itself, leading to \"gradient-of-a-gradient\" calculations with computationally expensive second-order and mixed derivatives. While modern automatic differentiation libraries provide a convenient way to write programs for calculating these derivatives, they oftentimes cannot fully exploit the specific structure of these problems out-of-the-box, leading to suboptimal performance. In this paper, we analyse such cases and propose Mixed-Flow Meta-Gradients, or MixFlow-MG -- a practical algorithm that uses mixed-mode differentiation to construct more efficient and scalable computational graphs yielding over 10x memory and up to 25\\% wall-clock time improvements over standard implementations in modern meta-learning setups."
},
{
"venue": "ICML",
"title": "PINNsAgent: Automated PDE Surrogation with Large Language Models",
"authors": [
"Qingpo Wuwu",
"Chonghan Gao",
"Tianyu Chen",
"Yihang Huang",
"Yuekai Zhang",
"Jianing Wang",
"Jianxin Li",
"Haoyi Zhou",
"Shanghang Zhang"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=RO5OGOzs6M",
"source": "openreview",
"doi": "",
"abstract": "Solving partial differential equations (PDEs) using neural methods has been a long-standing scientific and engineering research pursuit. Physics-Informed Neural Networks (PINNs) have emerged as a promising alternative to traditional numerical methods for solving PDEs. However, the gap between domain-specific knowledge and deep learning expertise often limits the practical application of PINNs. Previous works typically involve manually conducting extensive PINNs experiments and summarizing heuristic rules for hyperparameter tuning. In this work, we introduce PINNsAgent, a novel surrogation framework that leverages large language models (LLMs) to bridge the gap between domain-specific knowledge and deep learning. PINNsAgent integrates Physics-Guided Knowledge Replay (PGKR) for efficient knowledge transfer from solved PDEs to similar problems, and Memory Tree Reasoning for exploring the search space of optimal PINNs architectures. We evaluate PINNsAgent on 14 benchmark PDEs, demonstrating its effectiveness in automating the surrogation process and significantly improving the accuracy of PINNs-based solutions."
},
{
"venue": "ICML",
"title": "Compositional Generalization via Forced Rendering of Disentangled Latents",
"authors": [
"Qiyao Liang",
"Daoyuan Qian",
"Liu Ziyin",
"Ila R Fiete"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=rkHCHI5H5W",
"source": "openreview",
"doi": "",
"abstract": "Composition—the ability to generate myriad variations from finite means—is believed to underlie powerful generalization. However, compositional generalization remains a key challenge for deep learning. A widely held assumption is that learning disentangled (factorized) representations naturally supports this kind of extrapolation. Yet, empirical results are mixed, with many generative models failing to recognize and compose factors to generate out-of-distribution (OOD) samples. In this work, we investigate a controlled 2D Gaussian \"bump\" generation task with fully disentangled $(x,y)$ inputs, demonstrating that standard generative architectures still fail in OOD regions when training with partial data, by re-entangling latent representations in subsequent layers. By examining the model's learned kernels and manifold geometry, we show that this failure reflects a \"memorization\" strategy for generation via data superposition rather than via composition of the true factorized features. We show that when models are forced—through architectural modifications with regularization or curated training data—to render the disentangled latents into the full-dimensional representational (pixel) space, they can be highly data-efficient and effective at composing in OOD regions. These findings underscore that disentangled latents in an abstract representation are insufficient and show that if models can represent disentangled factors directly in the output representational space, it can achieve robust compositional generalization."
},
{
"venue": "ICML",
"title": "EcoMapper: Generative Modeling for Climate-Aware Satellite Imagery",
"authors": [
"Muhammed Goktepe",
"Amir hossein Shamseddin",
"Erencan Uysal",
"Javier Muinelo Monteagudo",
"Lukas Drees",
"Aysim Toker",
"Senthold Asseng",
"Malte von Bloh"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=YUtJsxQjv3",
"source": "openreview",
"doi": "",
"abstract": "Satellite imagery is essential for Earth observation, enabling applications like crop yield prediction, environmental monitoring, and climate\nchange assessment. However, integrating satellite imagery with climate data remains a challenge, limiting its utility for forecasting and scenario analysis. We introduce a novel dataset of 2.9 million Sentinel-2 images spanning 15 land cover types with corresponding climate records, forming the foundation for two satellite image generation approaches using fine-tuned Stable Diffusion 3 models. The first is a text-to-image generation model that uses textual prompts with climate and land cover details to produce realistic synthetic imagery for specific regions. The second leverages ControlNet for multi-conditional image generation, preserving spatial structures while mapping climate data or generating time-series to simulate landscape evolution. By combining synthetic image generation with climate and land cover data, our work advances generative modeling in remote sensing, offering realistic inputs for environmental forecasting and new possibilities for climate adaptation and geospatial analysis."
},
{
"venue": "ICML",
"title": "An Empirical Study on Configuring In-Context Learning Demonstrations for Unleashing MLLMs' Sentimental Perception Capability",
"authors": [
"Daiqing Wu",
"Dongbao Yang",
"Sicheng Zhao",
"Can Ma",
"Yu Zhou"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=DTdtM53iag",
"source": "openreview",
"doi": "",
"abstract": "The advancements in Multimodal Large Language Models (MLLMs) have enabled various multimodal tasks to be addressed under a zero-shot paradigm. This paradigm sidesteps the cost of model fine-tuning, emerging as a dominant trend in practical application. Nevertheless, Multimodal Sentiment Analysis (MSA), a pivotal challenge in the quest for general artificial intelligence, fails to accommodate this convenience. The zero-shot paradigm exhibits undesirable performance on MSA, casting doubt on whether MLLMs can perceive sentiments as competent as supervised models. By extending the zero-shot paradigm to In-Context Learning (ICL) and conducting an in-depth study on configuring demonstrations, we validate that MLLMs indeed possess such capability. Specifically, three key factors that cover demonstrations' retrieval, presentation, and distribution are comprehensively investigated and optimized. A sentimental predictive bias inherent in MLLMs is also discovered and later effectively counteracted. By complementing each other, the devised strategies for three factors result in average accuracy improvements of 15.9% on six MSA datasets against the zero-shot paradigm and 11.2% against the random ICL baseline."
},
{
"venue": "ICML",
"title": "Beyond Cropped Regions: New Benchmark and Corresponding Baseline for Chinese Scene Text Retrieval in Diverse Layouts",
"authors": [
"Gengluo Li",
"Huawen Shen",
"Yu Zhou"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=G80YGyxzv7",
"source": "openreview",
"doi": "",
"abstract": "Chinese scene text retrieval is a practical task that aims to search for images containing visual instances of a Chinese query text. This task is extremely challenging because Chinese text often features complex and diverse layouts in real-world scenes. Current efforts tend to inherit the solution for English scene text retrieval, failing to achieve satisfactory performance. In this paper, we establish a Diversified Layout benchmark for Chinese Street View Text Retrieval (DL-CSVTR), which is specifically designed to evaluate retrieval performance across various text layouts, including vertical, cross-line, and partial alignments. To address the limitations in existing methods, we propose Chinese Scene Text Retrieval CLIP (CSTR-CLIP), a novel model that integrates global visual information with multi-granularity alignment training. CSTR-CLIP applies a two-stage training process to overcome previous limitations, such as the exclusion of visual features outside the text region and reliance on single-granularity alignment, thereby enabling the model to effectively handle diverse text layouts. Experiments on existing benchmark show that CSTR-CLIP outperforms the previous state-of-the-art model by 18.82% accuracy and also provides faster inference speed. Further analysis on DL-CSVTR confirms the superior performance of CSTR-CLIP in handling various text layouts. The dataset and code will be publicly available to facilitate research in Chinese scene text retrieval."
},
{
"venue": "ICML",
"title": "Controlling Large Language Model with Latent Action",
"authors": [
"Chengxing Jia",
"Ziniu Li",
"Pengyuan Wang",
"Yi-Chen Li",
"Zhenyu Hou",
"Yuxiao Dong",
"Yang Yu"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=cEKrGCFXPA",
"source": "openreview",
"doi": "",
"abstract": "Adapting Large Language Models (LLMs) to downstream tasks using Reinforcement Learning (RL) has proven to be an effective approach. However, LLMs do not inherently define the structure of an agent for RL training, particularly in terms of specifying the action space. This paper studies learning a compact latent action space to enhance the controllability and exploration of RL for LLMs. Inspired by reinforcement learning from observations, we propose **Co**ntrolling Large Language Models with **L**atent **A**ctions **CoLA**, a framework that integrates a latent action space into pre-trained LLMs. **CoLA** employs an \\emph{inverse dynamics model} to extract latent actions conditioned on future tokens, ensuring that the next token prediction is partially influenced by these actions. Simultaneously, **CoLA** fine-tunes the pre-trained LLM to function as a \\emph{language world model}, capable of incorporating latent actions as inputs. Additionally, **CoLA** trains a \\emph{policy model} to generate actions within this language world model. The policy model can be trained via behavior cloning to mimic a standard language model or through RL to maximize task-specific rewards. In this work, we apply **CoLA** to the Llama-3.1-8B model. Our experiments demonstrate that, compared to RL with token-level actions, **CoLA**'s latent actions enable greater semantic diversity. For enhancing downstream tasks, we show that **CoLA** with RL achieves a score of 42.4 on the \\emph{math500} benchmark, surpassing the baseline score of 38.2, and reaches 68.2 when augmented with a Monte Carlo Tree Search variant. Furthermore, **CoLA** with RL consistently improves performance on agent-based tasks without degrading the pre-trained LLM's capabilities, unlike the baseline. Finally, **CoLA** reduces computation time by half in tasks involving enhanced thinking prompts for LLMs via RL. These results highlight **CoLA**'s potential to advance RL-based adaptation of LLMs for downstream applications. The CoLA model is available at \\url{https://huggingface.co/LAMDA-RL/Llama-3.1-CoLA-10B}."
},
{
"venue": "ICML",
"title": "Certification for Differentially Private Prediction in Gradient-Based Training",
"authors": [
"Matthew Robert Wicker",
"Philip Sosnin",
"Igor Shilov",
"Adrianna Janik",
"Mark Niklas Mueller",
"Yves-Alexandre de Montjoye",
"Adrian Weller",
"Calvin Tsay"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=viXwXCkA7N",
"source": "openreview",
"doi": "",
"abstract": "We study private prediction where differential privacy is achieved by adding noise to the outputs of a non-private model. Existing methods rely on noise proportional to the global sensitivity of the model, often resulting in sub-optimal privacy-utility trade-offs compared to private training. We introduce a novel approach for computing dataset-specific upper bounds on prediction sensitivity by leveraging convex relaxation and bound propagation techniques. By combining these bounds with the smooth sensitivity mechanism, we significantly improve the privacy analysis of private prediction compared to global sensitivity-based approaches. Experimental results across real-world datasets in medical image classification and natural language processing demonstrate that our sensitivity bounds are can be orders of magnitude tighter than global sensitivity. Our approach provides a strong basis for the development of novel privacy preserving technologies."
},
{
"venue": "ICML",
"title": "Multilayer Matrix Factorization via Dimension-Reducing Diffusion Variational Inference",
"authors": [
"Junbin Liu",
"Farzan Farnia",
"Wing-Kin Ma"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=Dd7Qo7TJpf",
"source": "openreview",
"doi": "",
"abstract": "Multilayer matrix factorization (MMF) has recently emerged as a generalized model of, and potentially a more expressive approach than, the classic matrix factorization.\nThis paper considers MMF under a probabilistic formulation, and our focus is on inference methods under variational inference.\nThe challenge in this context lies in determining a variational process that leads to a computationally efficient and accurate approximation of the maximum likelihood inference. \nOne well-known example is the variational autoencoder (VAE), which uses neural networks for the variational process.\nIn this work, we take insight from variational diffusion models in the context of generative models to develop variational inference for MMF.\nWe propose a dimension-reducing diffusion process that results in a new way to interact with the layered structures of the MMF model.\nExperimental results demonstrate that the proposed diffusion variational inference method leads to improved performance scores compared to several existing methods, including the VAE."
},
{
"venue": "ICML",
"title": "Compressing tree ensembles through Level-wise Optimization and Pruning",
"authors": [
"Laurens Devos",
"Timo Martens",
"Deniz Can Oruc",
"Wannes Meert",
"Hendrik Blockeel",
"Jesse Davis"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=9Klg7ce8D7",
"source": "openreview",
"doi": "",
"abstract": "Tree ensembles (e.g., gradient boosting decision trees) are often used in practice because they offer excellent predictive performance while still being easy and efficient to learn. In some contexts, it is important to additionally optimize their size: this is specifically the case when models need to have verifiable properties (verification of fairness, robustness, etc. is often exponential in the ensemble's size), or when models run on battery-powered devices (smaller ensembles consume less energy, increasing battery autonomy). For this reason, compression of tree ensembles is worth studying. This paper presents LOP, a method for compressing a given tree ensemble by pruning or entirely removing trees in it, while updating leaf predictions in such a way that predictive accuracy is mostly unaffected. Empirically, LOP achieves compression factors that are often 10 to 100 times better than that of competing methods."
},
{
"venue": "ICML",
"title": "Time Series Representations with Hard-Coded Invariances",
"authors": [
"Thibaut Germain",
"Chrysoula Kosma",
"Laurent Oudre"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=SaKPKyjDp6",
"source": "openreview",
"doi": "",
"abstract": "Automatically extracting robust representations from large and complex time series data is becoming imperative for several real-world applications. Unfortunately, the potential of common neural network architectures in capturing invariant properties of time series remains relatively underexplored. For instance, convolutional layers often fail to capture underlying patterns in time series inputs that encompass strong deformations, such as trends. Indeed, invariances to some deformations may be critical for solving complex time series tasks, such as classification, while guaranteeing good generalization performance.\nTo address these challenges, we mathematically formulate and technically design efficient and hard-coded *invariant convolutions* for specific group actions applicable to the case of time series.\nWe construct these convolutions by considering specific sets of deformations commonly observed in time series, including *scaling*, *offset shift*, and *trend*.\nWe further combine the proposed invariant convolutions with standard convolutions in single embedding layers, and we showcase the layer capacity to capture complex invariant time series properties in several scenarios."
},
{
"venue": "ICML",
"title": "Voronoi-grid-based Pareto Front Learning and Its Application to Collaborative Federated Learning",
"authors": [
"Mengmeng Chen",
"Xiaohu Wu",
"QIQI LIU",
"Tiantian He",
"Yew-Soon Ong",
"Yaochu Jin",
"Qicheng Lao",
"Han Yu"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=hrBfufwMzg",
"source": "openreview",
"doi": "",
"abstract": "Multi-objective optimization (MOO) exists extensively in machine learning, and aims to find a set of Pareto-optimal solutions, called the Pareto front, e.g., it is fundamental for multiple avenues of research in federated learning (FL). Pareto-Front Learning (PFL) is a powerful method implemented using Hypernetworks (PHNs) to approximate the Pareto front. This method enables the acquisition of a mapping function from a given preference vector to the solutions on the Pareto front. However, most existing PFL approaches still face two challenges: (a) sampling rays in high-dimensional spaces; (b) failing to cover the entire Pareto Front which has a convex shape. Here, we introduce a novel PFL framework, called as PHN-HVVS, which decomposes the design space into Voronoi grids and deploys a genetic algorithm (GA) for Voronoi grid partitioning within high-dimensional space. We put forward a new loss function, which effectively contributes to more extensive coverage of the resultant Pareto front and maximizes the HV Indicator. Experimental results on multiple MOO machine learning tasks demonstrate that PHN-HVVS outperforms the baselines significantly in generating Pareto front. Also, we illustrate that PHN-HVVS advances the methodologies of several recent problems in the FL field. The code is available at\nhttps://github.com/buptcmm/phnhvvs."
},
{
"venue": "ICML",
"title": "Optimal Sensor Scheduling and Selection for Continuous-Discrete Kalman Filtering with Auxiliary Dynamics",
"authors": [
"Mohamad Al Ahdab",
"John Leth",
"Zheng-Hua Tan"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=CAPNgWkEEk",
"source": "openreview",
"doi": "",
"abstract": "We study the Continuous-Discrete Kalman Filter (CD-KF) for State-Space Models (SSMs) where continuous-time dynamics are observed via multiple sensors with discrete, irregularly timed measurements. Our focus extends to scenarios in which the measurement process is coupled with the states of an auxiliary SSM. For instance, higher measurement rates may increase energy consumption or heat generation, while a sensor’s accuracy can depend on its own spatial trajectory or that of the measured target. Each sensor thus carries distinct costs and constraints associated with its measurement rate and additional constraints and costs on the auxiliary state. We model measurement occurrences as independent Poisson processes with sensor-specific rates and derive an upper bound on the mean posterior covariance matrix of the CD-KF along the mean auxiliary state. The bound is continuously differentiable with respect to the measurement rates, which enables efficient gradient-based optimization. Exploiting this bound, we propose a finite-horizon optimal control framework to optimize measurement rates and auxiliary-state dynamics jointly. We further introduce a deterministic method for scheduling measurement times from the optimized rates. Empirical results in state-space filtering and dynamic temporal Gaussian process regression demonstrate that our approach achieves improved trade-offs between resource usage and estimation accuracy."
},
{
"venue": "ICML",
"title": "Fairness on Principal Stratum: A New Perspective on Counterfactual Fairness",
"authors": [
"Haoxuan Li",
"Zeyu Tang",
"Zhichao Jiang",
"Zhuangyan Fang",
"Yue Liu",
"Zhi Geng",
"Kun Zhang"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=7jxa1o8rDW",
"source": "openreview",
"doi": "",
"abstract": "Fairness in human and algorithmic decision-making is crucial in areas such as criminal justice, education, and social welfare. Recently, counterfactual fairness has drawn increasing research interest, suggesting that decision-making for individuals should remain the same when intervening with different values on protected attributes. Nevertheless, the question of \"which attributes and individuals should be protected\" is rarely discussed in the existing counterfactual fairness literature. For example, when considering leg disability as a protected attribute, the algorithms should not treat individuals with leg disabilities differently in college admissions, but one may naturally consider this factor when selecting runner athletes. In other words, when and how to enforce fairness is expected to depend on the causal relation between the protected attribute and the outcome of interest. Formally, this paper proposes principal counterfactual fairness using the concept of principal stratification from the causal inference literature, focusing on whether an algorithm is counterfactually fair for individuals whose protected attribute has no individual causal effect on the outcome of interest. To examine whether an algorithm satisfies principal counterfactual fairness, we derive the statistical bounds and propose a post-processing approach to achieving principal counterfactual fairness with minimal individual decision changes. Experiments are conducted using synthetic and real-world datasets to verify the effectiveness of our methods."
},
{
"venue": "ICML",
"title": "Exponential Family Variational Flow Matching for Tabular Data Generation",
"authors": [
"Andrés Guzmán-Cordero",
"Floor Eijkelboom",
"Jan-Willem van de Meent"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=kjtvCSkSsy",
"source": "openreview",
"doi": "",
"abstract": "While denoising diffusion and flow matching have driven major advances in generative modeling, their application to tabular data remains limited, despite its ubiquity in real-world applications. \nTo this end, we develop *TabbyFlow*, a variational Flow Matching (VFM) method for tabular data generation. \nTo apply VFM to data with mixed continuous and discrete features, we introduce **Exponential Family Variational Flow Matching (EF-VFM)**, which represents heterogeneous data types using a general exponential family distribution. \nWe hereby obtain an efficient, data-driven objective based on moment matching, enabling principled learning of probability paths over mixed continuous and discrete variables. \nWe also establish a connection between variational flow matching and generalized flow matching objectives based on Bregman divergences. \nEvaluation on tabular data benchmarks demonstrates state-of-the-art performance compared to baselines."
},
{
"venue": "ICML",
"title": "LLMs Can Reason Faster Only If We Let Them",
"authors": [
"Bilgehan Sel",
"Lifu Huang",
"Naren Ramakrishnan",
"Ruoxi Jia",
"Ming Jin"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=uTv5rOPZr4",
"source": "openreview",
"doi": "",
"abstract": "Large language models (LLMs) are making inroads into classical AI problems such as automated planning, yet key shortcomings continue to hamper their integration. Chain-of-Thought (CoT) struggles in complex multi-step reasoning, and Tree-of-Thoughts requires multiple queries that increase computational overhead. Recently, Algorithm-of-Thoughts (AoT) have shown promise using in-context examples, at the cost of significantly longer solutions compared to CoT. Aimed at bridging the solution length gap between CoT and AoT, this paper introduces AoT-O3, which combines supervised finetuning on AoT-style plans with a reinforcement learning (RL) framework designed to reduce solution length. The RL component uses a reward model that favors concise, valid solutions while maintaining planning accuracy. Empirical evaluations indicate that AoT-O3 shortens solution length by up to 80\\% compared to baseline AoT while maintaining or surpassing prior performance. These findings suggest a promising pathway for more efficient, scalable LLM-based planning."
},
{
"venue": "ICML",
"title": "MAGELLAN: Metacognitive predictions of learning progress guide autotelic LLM agents in large goal spaces",
"authors": [
"Loris Gaven",
"Thomas Carta",
"Clément ROMAC",
"Cédric Colas",
"sylvain lamprier",
"Olivier Sigaud",
"Pierre-Yves Oudeyer"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=hRMAo5N66M",
"source": "openreview",
"doi": "",
"abstract": "Open-ended learning agents must efficiently prioritize goals in vast possibility spaces, focusing on those that maximize learning progress (LP). When such autotelic exploration is achieved by LLM agents trained with online RL in high-dimensional and evolving goal spaces, a key challenge for LP prediction is modeling one’s own competence, a form of metacognitive monitoring. Traditional approaches either require extensive sampling or rely on brittle expert-defined goal groupings. We introduce MAGELLAN, a metacognitive framework that lets LLM agents learn to predict their competence and learning progress online. By capturing semantic relationships between goals, MAGELLAN enables sample-efficient LP estimation and dynamic adaptation to evolving goal spaces through generalization. In an interactive learning environment, we show that MAGELLAN improves LP prediction efficiency and goal prioritization, being the only method allowing the agent to fully master a large and evolving goal space. These results demonstrate how augmenting LLM agents with a metacognitive ability for LP predictions can effectively scale curriculum learning to open-ended goal spaces."
},
{
"venue": "ICML",
"title": "The Harder Path: Last Iterate Convergence for Uncoupled Learning in Zero-Sum Games with Bandit Feedback",
"authors": [
"Côme Fiegel",
"Pierre Menard",
"Tadashi Kozuno",
"Michal Valko",
"Vianney Perchet"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=OmQcPgq9RN",
"source": "openreview",
"doi": "",
"abstract": "We study the problem of learning in zero-sum matrix games with repeated play and bandit feedback. \nSpecifically, we focus on developing uncoupled algorithms that guarantee, without communication between players, convergence of the last-iterate to a Nash equilibrium. Although the non-bandit case has been studied extensively, this setting has only been explored recently, with a bound of $\\mathcal{O}(T^{-1/8})$ on the exploitability gap. We show that, for uncoupled algorithms, guaranteeing convergence of the policy profiles to a Nash equilibrium is detrimental to the performances, with the best attainable rate being $\\mathcal{O}(T^{-1/4})$ in contrast to the usual $\\mathcal{O}(T^{-1/2})$ rate for convergence of the average iterates. We then propose two algorithms that achieve this optimal rate. The first algorithm leverages a straightforward tradeoff between exploration and exploitation, while the second employs a regularization technique based on a two-step mirror descent approach."
},
{
"venue": "ICML",
"title": "Towards Efficient Online Tuning of VLM Agents via Counterfactual Soft Reinforcement Learning",
"authors": [
"Lang Feng",
"Weihao Tan",
"Zhiyi Lyu",
"Longtao Zheng",
"Haiyang Xu",
"Ming Yan",
"Fei Huang",
"Bo An"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=H76PMm7hf2",
"source": "openreview",
"doi": "",
"abstract": "Online fine-tuning vision-language model (VLM) agents with reinforcement learning (RL) has shown promise for equipping agents with multi-step, goal-oriented capabilities in dynamic environments. However, their open-ended textual action space and non-end-to-end nature of action generation present significant challenges to effective online exploration in RL, e.g., explosion of the exploration space. We propose a novel online fine-tuning method, Counterfactual Soft Reinforcement Learning (CoSo), better suited to the textual output space of VLM agents. Compared to prior methods that assign uniform uncertainty to all tokens, CoSo leverages counterfactual reasoning to dynamically assess the causal influence of individual tokens on post-processed actions. By prioritizing the exploration of action-critical tokens while reducing the impact of semantically redundant or low-impact tokens, CoSo enables a more targeted and efficient online rollout process. We provide theoretical analysis proving CoSo's convergence and policy improvement guarantees, and extensive empirical evaluations supporting CoSo's effectiveness. Our results across a diverse set of agent tasks, including Android device control, card gaming, and embodied AI, highlight its remarkable ability to enhance exploration efficiency and deliver consistent performance gains. The code is available at https://github.com/langfengQ/CoSo."
},
{
"venue": "ICML",
"title": "Neural Guided Diffusion Bridges",
"authors": [
"Gefan Yang",
"Frank van der Meulen",
"Stefan Sommer"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=4LClOWTAth",
"source": "openreview",
"doi": "",
"abstract": "We propose a novel method for simulating conditioned diffusion processes (diffusion bridges) in Euclidean spaces. By training a neural network to approximate bridge dynamics, our approach eliminates the need for computationally intensive Markov Chain Monte Carlo (MCMC) methods or reverse-process modeling. Compared to existing methods, it offers greater robustness across various diffusion specifications and conditioning scenarios. This applies in particular to rare events and multimodal distributions, which pose challenges for score-learning- and MCMC-based approaches. We propose a flexible variational family for approximating the diffusion bridge path measure which is partially specified by a neural network. Once trained, it enables efficient independent sampling at a cost comparable to sampling the unconditioned (forward) process."
},
{
"venue": "ICML",
"title": "ELMO : Efficiency via Low-precision and Peak Memory Optimization in Large Output Spaces",
"authors": [
"Jinbin Zhang",
"Nasib Ullah",
"Erik Schultheis",
"Rohit Babbar"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=d6CTIPrTTC",
"source": "openreview",
"doi": "",
"abstract": "Large output spaces, also referred to as Extreme multilabel classification (XMC), is a setting that arises, e.g., in large-scale tagging and product-to-product recommendation, and is characterized by the number of labels ranging from hundreds of thousands to millions. This means that the linear classification head, usually only a tiny fraction of the overall model, turns into the main driver for compute and memory demand. Current state-of-the-art XMC methods predominantly rely on FP16-FP32 mixed-precision training, which we show can be unstable, and inefficient in terms of memory usage and computational overhead. Meanwhile, existing low-precision methods typically retain higher precision for the classification layer. In this work, we propose ELMO, a pure low-precision training framework for XMC models using BFloat16 and Float8 data types. By leveraging Kahan summation and stochastic rounding, we demonstrate that XMC models can be effectively trained entirely in Float8, without relying on single-precision master weights or tensor scaling. Low-precision training, combined with our proposed memory optimizations---gradient fusion and chunking---enables significant reductions in GPU memory usage. For example, we train a 3-million-label XMC model with only 6.6 GiB of GPU memory, compared to the 39.7GiB required by the optimized SOTA method, Renee without compromising accuracy."
},
{
"venue": "ICML",
"title": "Hyper-Transforming Latent Diffusion Models",
"authors": [
"Ignacio Peis",
"Batuhan Koyuncu",
"Isabel Valera",
"Jes Frellsen"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=yhgcRwJ9Dn",
"source": "openreview",
"doi": "",
"abstract": "We introduce a novel generative framework for functions by integrating Implicit Neural Representations (INRs) and Transformer-based hypernetworks into latent variable models. Unlike prior approaches that rely on MLP-based hypernetworks with scalability limitations, our method employs a Transformer-based decoder to generate INR parameters from latent variables, addressing both representation capacity and computational efficiency. Our framework extends latent diffusion models (LDMs) to INR generation by replacing standard decoders with a Transformer-based hypernetwork, which can be trained either from scratch or via hyper-transforming—a strategy that fine-tunes only the decoder while freezing the pre-trained latent space. This enables efficient adaptation of existing generative models to INR-based representations without requiring full retraining. We validate our approach across multiple modalities, demonstrating improved scalability, expressiveness, and generalization over existing INR-based generative models. Our findings establish a unified and flexible framework for learning structured function representations."
},
{
"venue": "ICML",
"title": "Explicit Preference Optimization: No Need for an Implicit Reward Model",
"authors": [
"Xiangkun Hu",
"Lemin Kong",
"Tong He",
"David Wipf"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=iXvm0zvspb",
"source": "openreview",
"doi": "",
"abstract": "The generated responses of large language models (LLMs) are often fine-tuned to human preferences through a process called reinforcement learning from human feedback (RLHF). As RLHF relies on a challenging training sequence, whereby a separate reward model is independently learned and then later applied to LLM policy updates, ongoing research effort has targeted more straightforward alternatives. In this regard, direct preference optimization (DPO) and its many offshoots circumvent the need for a separate reward training step. Instead, through the judicious use of a reparameterization trick that induces an implicit reward, DPO and related methods consolidate learning to the minimization of a single loss function. And yet despite demonstrable success in some real-world settings, we prove that DPO-based objectives are nonetheless subject to sub-optimal regularization and counter-intuitive interpolation behaviors, underappreciated artifacts of the reparameterizations upon which they are based. To this end, we introduce an explicit preference optimization framework termed EXPO that requires no analogous reparameterization to achieve an implicit reward. Quite differently, we merely posit intuitively-appealing regularization factors from scratch that transparently avoid the potential pitfalls of key DPO variants, provably satisfying regularization desiderata that prior methods do not. Empirical results serve to corroborate our analyses and showcase the efficacy of EXPO."
},
{
"venue": "ICML",
"title": "CaDA: Cross-Problem Routing Solver with Constraint-Aware Dual-Attention",
"authors": [
"Han Li",
"Fei Liu",
"Zhi Zheng",
"Yu Zhang",
"Zhenkun Wang"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=CS4RyQuTig",
"source": "openreview",
"doi": "",
"abstract": "Vehicle routing problems (VRPs) are significant combinatorial optimization problems (COPs) holding substantial practical importance. Recently, neural combinatorial optimization (NCO), which involves training deep learning models on extensive data to learn vehicle routing heuristics, has emerged as a promising approach due to its efficiency and the reduced need for manual algorithm design. However, applying NCO across diverse real-world scenarios with various constraints necessitates cross-problem capabilities. Current cross-problem NCO methods for VRPs typically employ a constraint-unaware model, limiting their cross-problem performance. Furthermore, they rely solely on global connectivity, which fails to focus on key nodes and leads to inefficient representation learning. This paper introduces a \\underline{C}onstraint-\\underline{A}ware \\underline{D}ual-\\underline{A}ttention Model (CaDA), designed to address these limitations. CaDA incorporates a constraint prompt that efficiently represents different problem variants. Additionally, it features a dual-attention mechanism with a global branch for capturing broader graph-wide information and a sparse branch that selectively focuses on the key node connections. We comprehensively evaluate our model on 16 different VRPs and compare its performance against existing cross-problem VRP solvers. CaDA achieves state-of-the-art results across all tested VRPs. Our ablation study confirms that each component contributes to its cross-problem learning performance. The source code for CaDA is publicly available at \\url{https://github.com/CIAM-Group/CaDA}."
},
{
"venue": "ICML",
"title": "Uncertainty Estimation for Heterophilic Graphs Through the Lens of Information Theory",
"authors": [
"Dominik Fuchsgruber",
"Tom Wollschläger",
"Johannes Bordne",
"Stephan Günnemann"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=GDvO6viRCF",
"source": "openreview",
"doi": "",
"abstract": "While uncertainty estimation for graphs recently gained traction, most methods rely on homophily and deteriorate in heterophilic settings.\n We address this by analyzing message passing neural networks from an information-theoretic perspective and developing a suitable analog to data processing inequality to quantify information throughout the model's layers. In contrast to non-graph domains, information about the node-level prediction target can *increase* with model depth if a node's features are semantically different from its neighbors. \n Therefore, on heterophilic graphs, the latent embeddings of an MPNN each provide different information about the data distribution - different from homophilic settings.\n This reveals that considering all node representations simultaneously is a key design principle for epistemic uncertainty estimation on graphs beyond homophily. \n We empirically confirm this with a simple post-hoc density estimator on the joint node embedding space that provides state-of-the-art uncertainty on heterophilic graphs. At the same time, it matches prior work on homophilic graphs without explicitly exploiting homophily through post-processing."
},
{
"venue": "ICML",
"title": "Near Optimal Best Arm Identification for Clustered Bandits",
"authors": [
"Yash",
"Avishek Ghosh",
"Nikhil Karamchandani"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=3Jr5Al16MS",
"source": "openreview",
"doi": "",
"abstract": "This work investigates the problem of best arm identification for multi-agent multi-armed bandits. We consider $N$ agents grouped into $M$ clusters, where each cluster solves a stochastic bandit problem. The mapping between agents and bandits is \\textit{a priori} unknown. Each bandit is associated with $K$ arms, and the goal is to identify the best arm for each agent under a $\\delta$-probably correct ($\\delta$-PC) framework, while minimizing sample complexity and communication overhead. We propose two novel algorithms: \\emph{Clustering then Best Arm Identification} (\\texttt{Cl-BAI}) and \\emph{Best Arm Identification then Clustering} (\\texttt{BAI-Cl}). \\texttt{Cl-BAI} employs a two-phase approach that first clusters agents based on the bandit problems they are learning, followed by identifying the best arm for each cluster. \\texttt{BAI-Cl} reverses the sequence by identifying the best arms first and then clustering agents accordingly. Both algorithms exploit the successive elimination framework to ensure computational efficiency and high accuracy. Theoretical analysis establishes $\\delta$-PC guarantees for both methods, derives bounds on their sample complexity, and provides a lower bound for the problem class. Moreover, when $M$ is small (a constant), we show that the sample complexity of (a variant of) \\texttt{BAI-Cl} is (order-wise) minimax optimal. Experiments on synthetic and real-world (Movie Lens, Yelp) data demonstrates the superior performance of the proposed algorithms in terms of sample and communication efficiency, particularly in settings where $M \\ll N$."
},
{
"venue": "ICML",
"title": "FOUNDER: Grounding Foundation Models in World Models for Open-Ended Embodied Decision Making",
"authors": [
"Yucen Wang",
"Rui Yu",
"Shenghua Wan",
"Le Gan",
"De-Chuan Zhan"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=UTT5OTyIWm",
"source": "openreview",
"doi": "",
"abstract": "Foundation Models (FMs) and World Models (WMs) offer complementary strengths in task generalization at different levels. In this work, we propose FOUNDER, a framework that integrates the generalizable knowledge embedded in FMs with the dynamic modeling capabilities of WMs to enable open-ended task solving in embodied environments in a reward-free manner. We learn a mapping function that grounds FM representations in the WM state space, effectively inferring the agent's physical states in the world simulator from external observations. This mapping enables the learning of a goal-conditioned policy through imagination during behavior learning, with the mapped task serving as the goal state. Our method leverages the predicted temporal distance to the goal state as an informative reward signal. FOUNDER demonstrates superior performance on various multi-task offline visual control benchmarks, excelling in capturing the deep-level semantics of tasks specified by text or videos, particularly in scenarios involving complex observations or domain gaps where prior methods struggle. The consistency of our learned reward function with the ground-truth reward is also empirically validated. Our project website is https://sites.google.com/view/founder-rl."
},
{
"venue": "ICML",
"title": "Understanding Sharpness Dynamics in NN Training with a Minimalist Example: The Effects of Dataset Difficulty, Depth, Stochasticity, and More",
"authors": [
"Geonhui Yoo",
"Minhak Song",
"Chulhee Yun"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=XfjrLEPOQV",
"source": "openreview",
"doi": "",
"abstract": "When training deep neural networks with gradient descent, sharpness often increases---a phenomenon known as *progressive sharpening*---before saturating at the *edge of stability*. Although commonly observed in practice, the underlying mechanisms behind progressive sharpening remain poorly understood. In this work, we study this phenomenon using a minimalist model: a deep linear network with a single neuron per layer. We show that this simple model effectively captures the sharpness dynamics observed in recent empirical studies, offering a simple testbed to better understand neural network training. Moreover, we theoretically analyze how dataset properties, network depth, stochasticity of optimizers, and step size affect the degree of progressive sharpening in the minimalist model. We then empirically demonstrate how these theoretical insights extend to practical scenarios. This study offers a deeper understanding of sharpness dynamics in neural network training, highlighting the interplay between depth, training data, and optimizers."
},
{
"venue": "ICML",
"title": "Comparing Few to Rank Many: Active Human Preference Learning Using Randomized Frank-Wolfe Method",
"authors": [
"Kiran Koshy Thekumparampil",
"Gaurush Hiranandani",
"Kousha Kalantari",
"Shoham Sabach",
"Branislav Kveton"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=cUNfm13VUR",
"source": "openreview",
"doi": "",
"abstract": "We study learning human preferences from limited comparison feedback, a core machine learning problem that is at the center of reinforcement learning from human feedback (RLHF). We formulate the problem as learning a Plackett-Luce (PL) model from a limited number of $K$-subset comparisons over a universe of $N$ items, where typically $K \\ll N$. Our objective is to select the $K$-subsets such that all items can be ranked with minimal mistakes within the budget. We solve the problem using the D-optimal design, which minimizes the worst-case ranking loss under the estimated PL model. All known algorithms for this problem are computationally infeasible in our setting because we consider exponentially many subsets in $K$. To address this challenge, we propose a randomized Frank-Wolfe algorithm with memoization and sparse updates that has a low $O(N^2 + K^2)$ per-iteration complexity. We analyze it and demonstrate its empirical superiority on synthetic and open-source NLP datasets."
},
{
"venue": "ICML",
"title": "Score-Based Diffusion Policy Compatible with Reinforcement Learning via Optimal Transport",
"authors": [
"Mingyang Sun",
"Pengxiang Ding",
"Weinan Zhang",
"Donglin Wang"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=2dqiqST8ZJ",
"source": "openreview",
"doi": "",
"abstract": "Diffusion policies have shown promise in learning complex behaviors from demonstrations, particularly for tasks requiring precise control and long-term planning. However, they face challenges in robustness when encountering distribution shifts. This paper explores improving diffusion-based imitation learning models through online interactions with the environment. We propose OTPR (Optimal Transport-guided score-based diffusion Policy for Reinforcement learning fine-tuning), a novel method that integrates diffusion policies with RL using optimal transport theory. OTPR leverages the Q-function as a transport cost and views the policy as an optimal transport map, enabling efficient and stable fine-tuning. Moreover, we introduce masked optimal transport to guide state-action matching using expert keypoints and a compatibility-based resampling strategy to enhance training stability. Experiments on three simulation tasks demonstrate OTPR's superior performance and robustness compared to existing methods, especially in complex and sparse-reward environments. In sum, OTPR provides an effective framework for combining IL and RL, achieving versatile and reliable policy learning."
},
{
"venue": "ICML",
"title": "Accelerating Linear Recurrent Neural Networks for the Edge with Unstructured Sparsity",
"authors": [
"Alessandro Pierro",
"Steven Abreu",
"Jonathan Timcheck",
"Philipp Stratmann",
"Andreas Wild",
"Sumit Bam Shrestha"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=UNrfYfbLZ3",
"source": "openreview",
"doi": "",
"abstract": "Linear recurrent neural networks enable powerful long-range sequence modeling with constant memory usage and time-per-token during inference. These architectures hold promise for streaming applications at the edge, but deployment in resource-constrained environments requires hardware-aware optimizations to minimize latency and energy consumption. \nUnstructured sparsity offers a compelling solution, enabling substantial reductions in compute and memory requirements--when accelerated by compatible hardware platforms. \nIn this paper, we conduct a scaling study to investigate the Pareto front of performance and efficiency across inference compute budgets.\nWe find that highly sparse linear RNNs *consistently* achieve better efficiency-performance trade-offs than dense baselines, with $2\\times$ less compute and $36$\\% less memory at iso-accuracy.\nOur models achieve state-of-the-art results on a real-time streaming task for audio denoising.\nBy quantizing our sparse models to fixed-point arithmetic and deploying them on the Intel Loihi 2 neuromorphic chip for real-time processing, we translate model compression into tangible gains of $42\\times$ lower latency and $149\\times$ lower energy consumption compared to a dense model on an edge GPU.\nOur findings showcase the transformative potential of unstructured sparsity, paving the way for highly efficient recurrent neural networks in real-world, resource-constrained environments."
},
{
"venue": "ICML",
"title": "Complex Wavelet Mutual Information Loss: A Multi-Scale Loss Function for Semantic Segmentation",
"authors": [
"Renhao Lu"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=iwkCnlOa2A",
"source": "openreview",
"doi": "",
"abstract": "Recent advancements in deep neural networks have significantly enhanced the performance of semantic segmentation. However, class imbalance and instance imbalance remain persistent challenges, where smaller instances and thin boundaries are often overshadowed by larger structures. To address the multiscale nature of segmented objects, various models have incorporated mechanisms such as spatial attention and feature pyramid networks. Despite these advancements, most loss functions are still primarily pixel-wise, while regional and boundary-focused loss functions often incur high computational costs or are restricted to small-scale regions. To address this limitation, we propose the complex wavelet mutual information (CWMI) loss, a novel loss function that leverages mutual information from subband images decomposed by a complex steerable pyramid. The complex steerable pyramid captures features across multiple orientations and preserves structural similarity across scales. Meanwhile, mutual information is well-suited to capturing high-dimensional directional features and offers greater noise robustness. Extensive experiments on diverse segmentation datasets demonstrate that CWMI loss achieves significant improvements in both pixel-wise accuracy and topological metrics compared to state-of-the-art methods, while introducing minimal computational overhead. Our code is available at https://github.com/lurenhaothu/CWMI"
},
{
"venue": "ICML",
"title": "Contour Integration Underlies Human-Like Vision",
"authors": [
"Ben Lonnqvist",
"Elsa Scialom",
"Abdulkadir Gokce",
"Zehra Merchant",
"Michael Herzog",
"Martin Schrimpf"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=ftR9OuiUJA",
"source": "openreview",
"doi": "",
"abstract": "Despite the tremendous success of deep learning in computer vision, models still fall behind humans in generalizing to new input distributions. Existing benchmarks do not investigate the specific failure points of models by analyzing performance under many controlled conditions. Our study systematically dissects where and why models struggle with contour integration - a hallmark of human vision -- by designing an experiment that tests object recognition under various levels of object fragmentation. Humans (n=50) perform at high accuracy, even with few object contours present. This is in contrast to models which exhibit substantially lower sensitivity to increasing object contours, with most of the over 1,000 models we tested barely performing above chance. Only at very large scales ($\\sim5B$ training dataset size) do models begin to approach human performance. Importantly, humans exhibit an integration bias - a preference towards recognizing objects made up of directional fragments over directionless fragments. We find that not only do models that share this property perform better at our task, but that this bias also increases with model training dataset size, and training models to exhibit contour integration leads to high shape bias. Taken together, our results suggest that contour integration is a hallmark of object vision that underlies object recognition performance, and may be a mechanism learned from data at scale."
},
{
"venue": "ICML",
"title": "Towards Universal Offline Black-Box Optimization via Learning Language Model Embeddings",
"authors": [
"Rong-Xi Tan",
"Ming Chen",
"Ke Xue",
"Yao Wang",
"Yaoyuan Wang",
"Fu Sheng",
"Chao Qian"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=NOV32X1Rq3",
"source": "openreview",
"doi": "",
"abstract": "The pursuit of universal black-box optimization (BBO) algorithms is a longstanding goal. However, unlike domains such as language or vision, where scaling structured data has driven generalization, progress in offline BBO remains hindered by the lack of unified representations for heterogeneous numerical spaces. Thus, existing offline BBO approaches are constrained to single-task and fixed-dimensional settings, failing to achieve cross-domain universal optimization. Recent advances in language models (LMs) offer a promising path forward: their embeddings capture latent relationships in a unifying way, enabling universal optimization across different data types possible. In this paper, we discuss multiple potential approaches, including an end-to-end learning framework in the form of next-token prediction, as well as prioritizing the learning of latent spaces with strong representational capabilities. To validate the effectiveness of these methods, we collect offline BBO tasks and data from open-source academic works for training. Experiments demonstrate the universality and effectiveness of our proposed methods. Our findings suggest that unifying language model priors and learning string embedding space can overcome traditional barriers in universal BBO, paving the way for general-purpose BBO algorithms. The code is provided at https://github.com/lamda-bbo/universal-offline-bbo."
},
{
"venue": "ICML",
"title": "Hyperbolic-PDE GNN: Spectral Graph Neural Networks in the Perspective of A System of Hyperbolic Partial Differential Equations",
"authors": [
"Juwei Yue",
"Haikuo Li",
"Jiawei Sheng",
"Xiaodong Li",
"Taoyu Su",
"Tingwen Liu",
"Li Guo"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=UJXbcJ7qXB",
"source": "openreview",
"doi": "",
"abstract": "Graph neural networks (GNNs) leverage message passing mechanisms to learn the topological features of graph data. Traditional GNNs learns node features in a spatial domain unrelated to the topology, which can hardly ensure topological features. In this paper, we formulates message passing as a system of hyperbolic partial differential equations (hyperbolic PDEs), constituting a dynamical system that explicitly maps node representations into a particular solution space. This solution space is spanned by a set of eigenvectors describing the topological structure of graphs. Within this system, for any moment in time, a node features can be decomposed into a superposition of the basis of eigenvectors. This not only enhances the interpretability of message passing but also enables the explicit extraction of fundamental characteristics about the topological structure. Furthermore, by solving this system of hyperbolic partial differential equations, we establish a connection with spectral graph neural networks (spectral GNNs), serving as a message passing enhancement paradigm for spectral GNNs.We further introduce polynomials to approximate arbitrary filter functions. Extensive experiments demonstrate that the paradigm of hyperbolic PDEs not only exhibits strong flexibility but also significantly enhances the performance of various spectral GNNs across diverse graph tasks."
},
{
"venue": "ICML",
"title": "Mind the Gap: A Practical Attack on GGUF Quantization",
"authors": [
"Kazuki Egashira",
"Robin Staab",
"Mark Vero",
"Jingxuan He",
"Martin Vechev"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=TV17MLZGuA",
"source": "openreview",
"doi": "",
"abstract": "With the increasing size of frontier LLMs, post-training quantization has become the standard for memory-efficient deployment. Recent work has shown that basic rounding-based quantization schemes pose security risks, as they can be exploited to inject malicious behaviors into quantized models that remain hidden in full precision. However, existing attacks cannot be applied to more complex quantization methods, such as the GGUF family used in the popular ollama and llama.cpp frameworks. In this work, we address this gap by introducing the first attack on GGUF. Our key insight is that the quantization error -- the difference between the full-precision weights and their (de-)quantized version -- provides sufficient flexibility to construct malicious quantized models that appear benign in full precision. Leveraging this, we develop an attack that trains the target malicious LLM while constraining its weights based on quantization errors. We demonstrate the effectiveness of our attack on three popular LLMs across nine GGUF quantization data types on three diverse attack scenarios: insecure code generation ($\\Delta$=$88.7\\%$), targeted content injection ($\\Delta$=$85.0\\%$), and benign instruction refusal ($\\Delta$=$30.1\\%$). Our attack highlights that (1) the most widely used post-training quantization method is susceptible to adversarial interferences, and (2) the complexity of quantization schemes alone is insufficient as a defense."
},
{
"venue": "ICML",
"title": "Elucidating the design space of language models for image generation",
"authors": [
"Xuantong LIU",
"Shaozhe Hao",
"Xianbiao Qi",
"Tianyang Hu",
"Jun Wang",
"Rong Xiao",
"Yuan Yao"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=EIfCH9OgjR",
"source": "openreview",
"doi": "",
"abstract": "The success of large language models (LLMs) in text generation has inspired their application to image generation. However, existing methods either rely on specialized designs with inductive biases or adopt LLMs without fully exploring their potential in vision tasks. In this work, we systematically investigate the design space of LLMs for image generation and demonstrate that LLMs can achieve near state-of-the-art performance without domain-specific designs, simply by making proper choices in tokenization methods, modeling approaches, scan patterns, vocabulary design, and sampling strategies. We further analyze autoregressive models' learning and scaling behavior, revealing how larger models effectively capture more useful information than the smaller ones. Additionally, we explore the inherent differences between text and image modalities, highlighting the potential of LLMs across domains. The exploration provides valuable insights to inspire more effective designs when applying LLMs to other domains. With extensive experiments, our proposed model, **ELM** achieves an FID of 1.54 on 256$\\times$256 ImageNet and an FID of 3.29 on 512$\\times$512 ImageNet, demonstrating the powerful generative potential of LLMs in vision tasks."
},
{
"venue": "ICML",
"title": "Be a Goldfish: Forgetting Bad Conditioning in Sparse Linear Regression via Variational Autoencoders",
"authors": [
"Kuheli Pratihar",
"Debdeep Mukhopadhyay"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=aTQtGq7IyT",
"source": "openreview",
"doi": "",
"abstract": "Variational Autoencoders (VAEs), a class of latent-variable generative models, have seen extensive use in high-fidelity synthesis tasks, yet their loss landscape remains poorly understood. Prior theoretical works on VAE loss analysis have focused on their latent-space representational capabilities, both in the optimal and limiting cases. Although these insights have guided better VAE designs, they also often restrict VAEs to problem settings where classical algorithms, such as Principal Component Analysis (PCA), can trivially guarantee globally optimal solutions. In this work, we push the boundaries of our understanding of VAEs beyond these traditional regimes to tackle NP-hard sparse inverse problems, for which no classical algorithms exist. Specifically, we examine the nontrivial Sparse Linear Regression (SLR) problem of recovering optimal sparse inputs in the presence of an ill-conditioned design matrix having correlated features. We provably show that, under a linear encoder-decoder architecture incorporating the product of the SLR design matrix with a trainable, sparsity-promoting diagonal matrix, any minimum of VAE loss is guaranteed to be an optimal solution. This property is especially useful for identifying (a) a preconditioning factor that reduces the eigenvalue spread, and (b) the corresponding optimal sparse representation. Lastly, our empirical analysis with different types of design matrices validates these findings and even demonstrates a higher recovery rate at low sparsity where traditional algorithms fail. Overall, this work highlights the flexible nature of the VAE loss, which can be adapted to efficiently solve computationally hard problems under specific constraints."
},
{
"venue": "ICML",
"title": "Incremental Gradient Descent with Small Epoch Counts is Surprisingly Slow on Ill-Conditioned Problems",
"authors": [
"Yujun Kim",
"Jaeyoung Cha",
"Chulhee Yun"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=LiXD7mpjU0",
"source": "openreview",
"doi": "",
"abstract": "Recent theoretical results demonstrate that the convergence rates of permutation-based SGD (e.g., random reshuffling SGD) are faster than uniform-sampling SGD; however, these studies focus mainly on the large epoch regime, where the number of epochs $K$ exceeds the condition number $\\kappa$. In contrast, little is known when $K$ is smaller than $\\kappa$, and it is still a challenging open question whether permutation-based SGD can converge faster in this small epoch regime (Safran and Shamir, 2021). As a step toward understanding this gap, we study the naive deterministic variant, Incremental Gradient Descent (IGD), on smooth and strongly convex functions. Our lower bounds reveal that for the small epoch regime, IGD can exhibit surprisingly slow convergence even when all component functions are strongly convex. Furthermore, when some component functions are allowed to be nonconvex, we prove that the optimality gap of IGD can be significantly worse throughout the small epoch regime. Our analyses reveal that the convergence properties of permutation-based SGD in the small epoch regime may vary drastically depending on the assumptions on component functions. Lastly, we supplement the paper with tight upper and lower bounds for IGD in the large epoch regime."
},
{
"venue": "ICML",
"title": "Lightweight Dataset Pruning without Full Training via Example Difficulty and Prediction Uncertainty",
"authors": [
"Yeseul Cho",
"Baekrok Shin",
"Changmin Kang",
"Chulhee Yun"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=9rLxi2cnZC",
"source": "openreview",
"doi": "",
"abstract": "Recent advances in deep learning rely heavily on massive datasets, leading to substantial storage and training costs. Dataset pruning aims to alleviate this demand by discarding redundant examples. However, many existing methods require training a model with a full dataset over a large number of epochs before being able to prune the dataset, which ironically makes the pruning process more expensive than just training the model on the entire dataset. To overcome this limitation, we introduce the **Difficulty and Uncertainty-Aware Lightweight (DUAL)** score, which aims to identify important samples from the early training stage by considering both example difficulty and prediction uncertainty. To address a catastrophic accuracy drop at an extreme pruning ratio, we further propose a pruning ratio-adaptive sampling using Beta distribution.\nExperiments on various datasets and learning scenarios such as image classification with label noise and image corruption, and model architecture generalization demonstrate the superiority of our method over previous state-of-the-art (SOTA) approaches. Specifically, on ImageNet-1k, our method reduces the time cost for pruning to 66\\% compared to previous methods while achieving a SOTA 60\\% test accuracy at a 90\\% pruning ratio. On CIFAR datasets, the time cost is reduced to just 15\\% while maintaining SOTA performance."
},
{
"venue": "ICML",
"title": "The Case for Learned Provenance-based System Behavior Baseline",
"authors": [
"Yao Zhu",
"Zhenyuan LI",
"Yangyang Wei",
"Shouling Ji"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=SY4owu5BK6",
"source": "openreview",
"doi": "",
"abstract": "Provenance graphs describe data flows and causal dependencies of host activities, enabling to track the data propagation and manipulation throughout the systems, which provide a foundation for intrusion detection. However, these Provenance-based Intrusion Detection Systems (PIDSes) face significant challenges in storage, representation, and analysis, which impede the efficacy of machine learning models such as Graph Neural Networks (GNNs) in processing and learning from these graphs. This paper presents a novel learning-based anomaly detection method designed to efficiently embed and analyze large-scale provenance graphs. Our approach integrates dynamic graph processing with adaptive encoding, facilitating compact embeddings that effectively address out-of-vocabulary (OOV) elements and adapt to normality shifts in dynamic real-world environments. Subsequently, we incorporate this refined baseline into a tag-propagation framework for real-time detection. Our evaluation demonstrates the method's accuracy and adaptability in anomaly path mining, significantly advancing the state-of-the-art in handling and analyzing provenance graphs for anomaly detection."
},
{
"venue": "ICML",
"title": "Flexible, Efficient, and Stable Adversarial Attacks on Machine Unlearning",
"authors": [
"Zihan Zhou",
"Yang Zhou",
"Zijie Zhang",
"Lingjuan Lyu",
"Da Yan",
"Ruoming Jin",
"Dejing Dou"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=ba3sSfEnj1",
"source": "openreview",
"doi": "",
"abstract": "Machine unlearning (MU) aims to remove the influence of specific data points from trained models, enhancing compliance with privacy regulations. However, the vulnerability of basic MU models to malicious unlearning requests in adversarial learning environments has been largely overlooked. Existing adversarial MU attacks suffer from three key limitations: inflexibility due to pre-defined attack targets, inefficiency in handling multiple attack requests, and instability caused by non-convex loss functions. To address these challenges, we propose a Flexible, Efficient, and Stable Attack (DDPA). First, leveraging Carathéodory's theorem, we introduce a convex polyhedral approximation to identify points in the loss landscape where convexity approximately holds, ensuring stable attack performance. Second, inspired by simplex theory and John's theorem, we develop a regular simplex detection technique that maximizes coverage over the parameter space, improving attack flexibility and efficiency. We theoretically derive the proportion of the effective parameter space occupied by the constructed simplex. We evaluate the attack success rate of our DDPA method on real datasets against state-of-the-art machine unlearning attack methods. Our source code is available at https://github.com/zzz0134/DDPA."
},
{
"venue": "ICML",
"title": "PipeOffload: Improving Scalability of Pipeline Parallelism with Memory Optimization",
"authors": [
"Xinyi Wan",
"Penghui Qi",
"Guangxing Huang",
"Min Lin",
"Jialin Li"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=O0lxLP4ABD",
"source": "openreview",
"doi": "",
"abstract": "Pipeline parallelism (PP) is widely used for training large language models (LLMs), yet its scalability is often constrained by high activation memory consumption as the number of in-flight microbatches grows with the degree of PP. In this paper, we focus on addressing this challenge by leveraging the under-explored memory offload strategy in PP. With empirical study, we discover that in the majority of standard configurations, at least half, and potentially all, of the activations can be offloaded with negligible overhead. In the cases where full overload is not possible, we introduce a novel selective offload strategy that decreases peak activation memory in a better-than-linear manner. Furthermore, we integrate memory offload with other techniques to jointly consider overall throughput and memory limitation. Our experiments proves that the per-device activation memory effectively reduces with the total number of stages, making PP a stronger alternative than TP, offering up to a 19\\% acceleration with even lower memory consumption."
},
{
"venue": "ICML",
"title": "Unified Screening for Multiple Diseases",
"authors": [
"Yiğit Narter",
"Alihan Hüyük",
"Mihaela van der Schaar",
"Cem Tekin"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=z4XS0Ie391",
"source": "openreview",
"doi": "",
"abstract": "Current screening programs that focus on improving patient health while minimizing screening costs are tailored for individual diseases. Designing unified screening programs for multiple diseases requires carefully balancing competing disease risks, which is an open problem. In this work, we address this problem by casting unified screening as a referral problem, in which we choose to activate a subset of screening policies for individual diseases by accounting for competing risks that influence patient outcomes. We introduce a novel optimization framework that incorporates disease risks, budget constraints, and diagnostic error limits and characterize the structural properties of the optimal referral policy. For the unified screening of two diseases, we show that the optimal activation threshold for the screening of one disease depends on the risk of the other, resulting in decision boundaries with distinct risk-dependent profiles. We compare our unified model with independent screening programs that apply isolated activation thresholds for screening of each disease. Our approach optimizes screening decisions collectively, improving overall survival outcomes, particularly for patients with high disease risks."
},
{
"venue": "ICML",
"title": "STAMP Your Content: Proving Dataset Membership via Watermarked Rephrasings",
"authors": [
"Saksham Rastogi",
"Pratyush Maini",
"Danish Pruthi"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=qF6mxani2X",
"source": "openreview",
"doi": "",
"abstract": "Given how large parts of publicly available text are crawled to pretrain large language models (LLMs), data creators increasingly worry about the inclusion of their proprietary data for model training without attribution or licensing. Their concerns are also shared by benchmark curators whose test-sets might be compromised. In this paper, we present STAMP, a framework for detecting dataset membership—i.e., determining the inclusion of a dataset in the pretraining corpora of LLMs. Given an original piece of content, our proposal involves first generating multiple rephrases, each embedding a watermark with a unique secret key. One version is to be released publicly, while others are to be kept private. Subsequently, creators can compare model likelihoods between public and private versions using paired statistical tests to prove membership. We show that our framework can successfully detect contamination across four benchmarks which appear only once in the training data and constitute less than 0.001% of the total tokens, outperforming several contamination detection and dataset inference baselines. We verify that STAMP preserves both the semantic meaning and utility of the original data. We apply STAMP to two real-world scenarios to confirm the inclusion of paper abstracts and blog articles in the pretraining corpora."
},
{
"venue": "ICML",
"title": "A Mixed-Curvature based Pre-training Paradigm for Multi-Task Vehicle Routing Solver",
"authors": [
"Suyu Liu",
"Zhiguang Cao",
"Shanshan Feng",
"Yew-Soon Ong"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=JsPyLqCgks",
"source": "openreview",
"doi": "",
"abstract": "Solving various types of vehicle routing problems (VRPs) using a unified neural solver has garnered significant attentions in recent years. Despite their effectiveness, existing neural multi-task solvers often fail to account for the geometric structures inherent in different tasks, which may result in suboptimal performance. To address this limitation, we propose a curvature-aware pre-training framework. Specifically, we leverage mixed-curvature spaces during the feature fusion stage, encouraging the model to capture the underlying geometric properties of each instance. Through extensive experiments, we evaluate the proposed pre-training strategy on existing neural multi-task solvers across a variety of testing scenarios. The results demonstrate that the curvature-aware pre-training approach not only enhances the generalization capabilities of existing neural VRP solvers on synthetic datasets but also improves solution quality on real-world benchmarks."
},
{
"venue": "ICML",
"title": "GRAM: A Generative Foundation Reward Model for Reward Generalization",
"authors": [
"Chenglong Wang",
"Yang Gan",
"Yifu Huo",
"Yongyu Mu",
"Qiaozhi He",
"MuRun Yang",
"Bei Li",
"Tong Xiao",
"Chunliang Zhang",
"Tongran Liu",
"JingBo Zhu"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=rxKC8v2uHc",
"source": "openreview",
"doi": "",
"abstract": "In aligning large language models (LLMs), reward models have played an important role, but are standardly trained as discriminative models and rely only on labeled human preference data. In this paper, we explore methods that train reward models using both unlabeled and labeled data. Building on the generative models in LLMs, we develop a generative reward model that is first trained via large-scale unsupervised learning and then fine-tuned via supervised learning. We also show that by using label smoothing, we are in fact optimizing a regularized pairwise ranking loss. This result, in turn, provides a new view of training reward models, which links generative models and discriminative models under the same class of training objectives. The outcome of these techniques is a foundation reward model, which can be applied to a wide range of tasks with little or no further fine-tuning effort. Extensive experiments show that this model generalizes well across several tasks, including response ranking, reinforcement learning from human feedback, and task adaptation with fine-tuning, achieving significant performance improvements over several strong baseline models."
},
{
"venue": "ICML",
"title": "Fine-Grained Captioning of Long Videos through Scene Graph Consolidation",
"authors": [
"Sanghyeok Chu",
"Seonguk Seo",
"Bohyung Han"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=aTC2euLwnh",
"source": "openreview",
"doi": "",
"abstract": "Recent advances in vision-language models have led to impressive progress in caption generation for images and short video clips. However, these models remain constrained by their limited temporal receptive fields, making it difficult to produce\ncoherent and comprehensive captions for long videos. While several methods have been proposed to aggregate information across video segments, they often rely on supervised fine-tuning or incur significant computational overhead. To address these challenges, we introduce a novel framework for long video captioning based on graph consolidation. Our approach first generates segment-level captions, corresponding to individual frames or short video intervals, using off-the-shelf visual captioning models. These captions are then parsed into individual scene graphs, which are subsequently consolidated into a unified graph representation that preserves both holistic context and fine-grained details throughout the video. A lightweight graph-to-text decoder then produces the final video-level caption. This framework effectively extends the temporal understanding capabilities of existing models without requiring any additional fine-tuning on long video datasets. Experimental results show that our method significantly outperforms existing LLM-based consolidation approaches, achieving strong zero-shot performance while substantially reducing computational costs."
},
{
"venue": "ICML",
"title": "M3-JEPA: Multimodal Alignment via Multi-gate MoE based on the Joint-Embedding Predictive Architecture",
"authors": [
"Hongyang Lei",
"Xiaolong Cheng",
"Qi Qin",
"Dan Wang",
"Huazhen Huang",
"Qingqing Gu",
"Yetao Wu",
"Luo Ji"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=tYwKQMMjJA",
"source": "openreview",
"doi": "",
"abstract": "Current multimodal learning strategies primarily optimize in the original token space. Such a framework is easy to incorporate with the backbone of pretrained language model, but might result in modality collapse. To alleviate such issues, we leverage the Joint-Embedding Predictive Architecture (JEPA) on the multimodal tasks, which converts the input embedding into the output embedding space by a predictor and then conducts the cross-modal alignment on the latent space. We implement this predictor by a Multi-Gate Mixture of Experts (MMoE) and name the framework as M3-JEPA, accordingly. The gating function disentangles the modality-specific and shared information and derives information-theoretic optimality. The framework is implemented with both contrastive and regularization loss, and solved by alternative gradient descent (AGD) between different multimodal tasks. By thoroughly designed experiments, we show that M3-JEPA can obtain state-of-the-art performance on different modalities and tasks, generalize to unseen datasets and domains, and is computationally efficient in both training and inference. Our observation suggests that M3-JEPA might become a new basis to self-supervised learning in the open world."
},
{
"venue": "ICML",
"title": "AEQA-NAT : Adaptive End-to-end Quantization Alignment Training Framework for Non-autoregressive Machine Translation",
"authors": [
"Xiangyu Qu",
"guojing liu",
"Liang Li"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=mQE0EsrX1y",
"source": "openreview",
"doi": "",
"abstract": "Non-autoregressive Transformers (NATs) have garnered significant attention due to their efficient decoding compared to autoregressive methods. However, existing conditional dependency modeling schemes based on masked language modeling introduce a *training-inference gap* in NATs. For instance, while NATs sample target words during training to enhance input, this condition cannot be met during inference, and simply annealing the sampling rate to zero during training leads to model performance degradation. We demonstrate that this *training-inference gap* prevents NATs from fully realizing their potential. To address this, we propose an adaptive end-to-end quantization alignment training framework, which introduces a semantic consistency space to adaptively align NAT training, eliminating the need for target information and thereby bridging the *training-inference gap*.Experimental results demonstrate that our method outperforms most existing fully NAT models, delivering performance on par with Autoregressive Transformer (AT) while being 17.0 times more efficient in inference."
},
{
"venue": "ICML",
"title": "Comparing Comparisons: Informative and Easy Human Feedback with Distinguishability Queries",
"authors": [
"Xuening Feng",
"Zhaohui JIANG",
"Timo Kaufmann",
"Eyke Hüllermeier",
"Paul Weng",
"Yifei Zhu"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=Cf8gsqWrua",
"source": "openreview",
"doi": "",
"abstract": "Learning human objectives from preference feedback has significantly advanced reinforcement learning (RL) in domains where objectives are hard to formalize. \nHowever, traditional methods based on pairwise trajectory comparisons face notable challenges, including the difficulty in comparing trajectories with subtle differences and the limitation of conveying only ordinal information, limiting direct inference of preference strength. \nIn this paper, we introduce a novel *distinguishability query*, enabling humans to express preference strength by comparing two pairs of trajectories. \nLabelers first indicate which of two pairs is easier to distinguish, then provide preference feedback only on the easier pair. \nOur proposed query type directly captures preference strength and is expected to reduce the cognitive load on the labeler. \nWe further connect this query to cardinal utility and difference relations and develop an efficient query selection scheme to achieve a better trade-off between query informativeness and easiness. \nExperimental results demonstrate the potential of our method for faster, data-efficient learning and improved user-friendliness in RLHF benchmarks, particularly in classical control settings where preference strength is critical for expected utility maximization."
},
{
"venue": "ICML",
"title": "BounDr.E: Predicting Drug-likeness via Biomedical Knowledge Alignment and EM-like One-Class Boundary Optimization",
"authors": [
"Dongmin Bang",
"Inyoung Sung",
"Yinhua Piao",
"Sangseon Lee",
"Sun Kim"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=Z9Xugry05b",
"source": "openreview",
"doi": "",
"abstract": "The advent of generative AI now enables large-scale $\\textit{de novo}$ design of molecules, but identifying viable drug candidates among them remains an open problem. Existing drug-likeness prediction methods often rely on ambiguous negative sets or purely structural features, limiting their ability to accurately classify drugs from non-drugs. In this work, we introduce BounDr.E}: a novel modeling of drug-likeness as a compact space surrounding approved drugs through a dynamic one-class boundary approach. Specifically, we enrich the chemical space through biomedical knowledge alignment, and then iteratively tighten the drug-like boundary by pushing non-drug-like compounds outside via an Expectation-Maximization (EM)-like process. Empirically, BounDr.E achieves 10\\% F1-score improvement over the previous state-of-the-art and demonstrates robust cross-dataset performance, including zero-shot toxic compound filtering. Additionally, we showcase its effectiveness through comprehensive case studies in large-scale $\\textit{in silico}$ screening. Our codes and constructed benchmark data under various schemes are provided at: https://github.com/eugenebang/boundr_e."
},
{
"venue": "ICML",
"title": "Provable Benefit of Random Permutations over Uniform Sampling in Stochastic Coordinate Descent",
"authors": [
"Donghwa Kim",
"Jaewook Lee",
"Chulhee Yun"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=KBUSuiLBMq",
"source": "openreview",
"doi": "",
"abstract": "We analyze the convergence rates of two popular variants of coordinate descent (CD): random CD (RCD), in which the coordinates are sampled uniformly at random, and random-permutation CD (RPCD), in which random permutations are used to select the update indices. Despite abundant empirical evidence that RPCD outperforms RCD in various tasks, the theoretical gap between the two algorithms’ performance has remained elusive. Even for the benign case of positive-definite quadratic functions with permutation-invariant Hessians, previous efforts have failed to demonstrate a provable performance gap between RCD and RPCD. To this end, we present novel results showing that, for a class of quadratics with permutation-invariant structures, the contraction rate upper bound for RPCD is always strictly smaller than the contraction rate lower bound for RCD for every individual problem instance. Furthermore, we conjecture that this function class contains the worst-case examples of RPCD among all positive-definite quadratics. Combined with our RCD lower bound, this conjecture extends our results to the general class of positive-definite quadratic functions."
},
{
"venue": "ICML",
"title": "Optimal and Practical Batched Linear Bandit Algorithm",
"authors": [
"Sanghoon Yu",
"Min-hwan Oh"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=WcFLasjwXs",
"source": "openreview",
"doi": "",
"abstract": "We study the linear bandit problem under limited adaptivity, known as the batched linear bandit. While existing approaches can achieve near-optimal regret in theory, they are often computationally prohibitive or underperform in practice. We propose BLAE, a novel batched algorithm that integrates arm elimination with regularized G-optimal design, achieving the minimax optimal regret (up to logarithmic factors in $T$) in both large-$K$ and small-$K$ regimes for the first time, while using only $O(\\log\\log T)$ batches. Our analysis introduces new techniques for batch-wise optimal design and refined concentration bounds. Crucially, BLAE demonstrates low computational overhead and strong empirical performance, outperforming state-of-the-art methods in extensive numerical evaluations. Thus, BLAE is the first algorithm to combine provable minimax-optimality in all regimes and practical superiority in batched linear bandits."
},
{
"venue": "ICML",
"title": "OR-Bench: An Over-Refusal Benchmark for Large Language Models",
"authors": [
"Justin Cui",
"Wei-Lin Chiang",
"Ion Stoica",
"Cho-Jui Hsieh"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=CdFnEu0JZV",
"source": "openreview",
"doi": "",
"abstract": "Large Language Models (LLMs) require careful safety alignment to prevent malicious outputs. While significant research focuses on mitigating harmful content generation, \nthe enhanced safety often come with the side effect of over-refusal, where LLMs may reject innocuous prompts and become less helpful.\nAlthough the issue of over-refusal has been empirically observed, a systematic measurement is challenging \ndue to the difficulty of crafting prompts that can elicit the over-refusal behaviors of LLMs.\nThis study proposes a novel method for automatically generating large-scale over-refusal datasets. Leveraging this technique, we introduce OR-Bench, the first large-scale over-refusal benchmark. OR-Bench comprises 80,000 over-refusal prompts across 10 common rejection categories, a subset of around 1,000 hard prompts that are challenging even for state-of-the-art LLMs, and an additional 600 toxic prompts to prevent indiscriminate responses.\nWe then conduct a comprehensive study to measure the over-refusal of 32 popular LLMs across 8 model families. Our datasets are publicly available at https://huggingface.co/bench-llms and our codebase is open-sourced at https://github.com/justincui03/or-bench.\n We hope this benchmark can help the community develop better safety aligned models."
},
{
"venue": "ICML",
"title": "TRACE Back from the Future: A Probabilistic Reasoning Approach to Controllable Language Generation",
"authors": [
"Gwen Yidou Weng",
"Benjie Wang",
"Guy Van den Broeck"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=LhkSfpfRXW",
"source": "openreview",
"doi": "",
"abstract": "As large language models (LMs) advance, there is an increasing need to control their outputs to align with human values (e.g., detoxification) or desired attributes (e.g., personalization, topic). However, autoregressive models focus on next-token predictions and struggle with global properties that require looking ahead. Existing solutions either post-train LMs for each new attribute—expensive and inflexible—or approximate the Expected Attribute Probability (EAP) of future sequences by sampling or training, which is slow and unreliable for rare attributes. We introduce **TRACE** (Tractable Probabilistic Reasoning for Adaptable Controllable gEneration), a novel framework that efficiently computes EAP and adapts to new attributes through tractable *probabilistic* reasoning and lightweight *control*. TRACE distills a Hidden Markov Model (HMM) from an LM and pairs it with a small classifier to estimate attribute probabilities, enabling exact EAP computation over the HMM’s predicted futures. This EAP is then used to reweigh the LM’s next-token probabilities for globally compliant continuations. Empirically, TRACE achieves state-of-the-art detoxification results with only 20% decoding overhead, yields 76 low-resource personalized LMs within seconds, and seamlessly extends to composite attributes."
},
{
"venue": "ICML",
"title": "A Theoretical Justification for Asymmetric Actor-Critic Algorithms",
"authors": [
"Gaspard Lambrechts",
"Damien Ernst",
"Aditya Mahajan"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=F1yANMCnAn",
"source": "openreview",
"doi": "",
"abstract": "In reinforcement learning for partially observable environments, many successful algorithms have been developed within the asymmetric learning paradigm. This paradigm leverages additional state information available at training time for faster learning. Although the proposed learning objectives are usually theoretically sound, these methods still lack a precise theoretical justification for their potential benefits. We propose such a justification for asymmetric actor-critic algorithms with linear function approximators by adapting a finite-time convergence analysis to this setting. The resulting finite-time bound reveals that the asymmetric critic eliminates error terms arising from aliasing in the agent state."
},
{
"venue": "ICML",
"title": "Reducing Confounding Bias without Data Splitting for Causal Inference via Optimal Transport",
"authors": [
"Yuguang Yan",
"Zongyu Li",
"Haolin Yang",
"Zeqin Yang",
"Hao Zhou",
"Ruichu Cai",
"Zhifeng Hao"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=fd7ddFBNmP",
"source": "openreview",
"doi": "",
"abstract": "Causal inference seeks to estimate the effect given a treatment such as a medicine or the dosage of a medication. To reduce the confounding bias caused by the non-randomized treatment assignment, most existing methods reduce the shift between subpopulations receiving different treatments. However, these methods split limited training samples into smaller groups, which cuts down the number of samples in each group, while precise distribution estimation and alignment highly rely on a sufficient number of training samples. In this paper, we propose a distribution alignment paradigm without data splitting, which can be naturally applied in the settings of binary and continuous treatments. To this end, we characterize the confounding bias by considering different probability measures of the same set including all the training samples, and exploit the optimal transport theory to analyze the confounding bias and outcome estimation error. Based on this, we propose to learn balanced representations by reducing the bias between the marginal distribution and the conditional distribution of a treatment. As a result, data reduction caused by splitting is avoided, and the outcome prediction model trained on one treatment group can be generalized to the entire population. The experiments on both binary and continuous treatment settings demonstrate the effectiveness of our method."
},
{
"venue": "ICML",
"title": "Provably Near-Optimal Federated Ensemble Distillation with Negligible Overhead",
"authors": [
"Won-Jun Jang",
"Hyeon-Seo Park",
"Si-Hyeon Lee"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=6znPjYn11w",
"source": "openreview",
"doi": "",
"abstract": "Federated ensemble distillation addresses client heterogeneity by generating pseudo-labels for an unlabeled server dataset based on client predictions and training the server model using the pseudo-labeled dataset. The unlabeled server dataset can either be pre-existing or generated through a data-free approach. The effectiveness of this approach critically depends on the method of assigning weights to client predictions when creating pseudo-labels, especially in highly heterogeneous settings. Inspired by theoretical results from GANs, we propose a provably near-optimal weighting method that leverages client discriminators trained with a server-distributed generator and local datasets. Our experiments on various image classification tasks demonstrate that the proposed method significantly outperforms baselines. Furthermore, we show that the additional communication cost, client-side privacy leakage, and client-side computational overhead introduced by our method are negligible, both in scenarios with and without a pre-existing server dataset."
},
{
"venue": "ICML",
"title": "Ensemble Distribution Distillation via Flow Matching",
"authors": [
"Jonggeon Park",
"Giung Nam",
"Hyunsu Kim",
"Jongmin Yoon",
"Juho Lee"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=waeJHU2oeI",
"source": "openreview",
"doi": "",
"abstract": "Neural network ensembles have proven effective in improving performance across a range of tasks; however, their high computational cost limits their applicability in resource-constrained environments or for large models. Ensemble distillation, the process of transferring knowledge from an ensemble teacher to a smaller student model, offers a promising solution to this challenge. The key is to ensure that the student model is both cost-efficient and achieves performance comparable to the ensemble teacher. With this in mind, we propose a novel ensemble distribution distillation method, which leverages flow matching to effectively transfer the diversity from the ensemble teacher to the student model. Our extensive experiments demonstrate the effectiveness of our proposed method compared to existing ensemble distillation approaches."
},
{
"venue": "ICML",
"title": "Internal Causal Mechanisms Robustly Predict Language Model Out-of-Distribution Behaviors",
"authors": [
"Jing Huang",
"Junyi Tao",
"Thomas Icard",
"Diyi Yang",
"Christopher Potts"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=Ofa1cspTrv",
"source": "openreview",
"doi": "",
"abstract": "Interpretability research now offers a variety of techniques for identifying abstract internal mechanisms in neural networks. Can such techniques be used to predict how models will behave on out-of-distribution examples? In this work, we provide a positive answer to this question. Through a diverse set of language modeling tasks—including symbol manipulation, knowledge retrieval, and instruction following—we show that the most robust features for correctness prediction are those that play a distinctive causal role in the model’s behavior. Specifically, we propose two methods that leverage causal mechanisms to predict the correctness of model outputs: counterfactual simulation (checking whether key causal variables are realized) and value probing (using the values of those variables to make predictions). Both achieve high AUC-ROC in distribution and outperform methods that rely on causal-agnostic features in out-of-distribution settings, where predicting model behaviors is more crucial. Our work thus highlights a novel and significant application for internal causal analysis of language models."
},
{
"venue": "ICML",
"title": "PDUDT: Provable Decentralized Unlearning under Dynamic Topologies",
"authors": [
"Jing Qiao",
"Yu Liu",
"Zengzhe Chen",
"Mingyi Li",
"YUAN YUAN",
"Xiao Zhang",
"Dongxiao Yu"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=K0Vg8b7nyI",
"source": "openreview",
"doi": "",
"abstract": "This paper investigates decentralized unlearning, aiming to eliminate the impact of a specific client on the whole decentralized system. However, decentralized communication characterizations pose new challenges for effective unlearning: the indirect connections make it difficult to trace the specific client's impact, while the dynamic topology limits the scalability of retraining-based unlearning methods.\nIn this paper, we propose the first **P**rovable **D**ecentralized **U**nlearning algorithm under **D**ynamic **T**opologies called PDUDT. It allows clients to eliminate the influence of a specific client without additional communication or retraining. We provide rigorous theoretical guarantees for PDUDT, showing it is statistically indistinguishable from perturbed retraining. Additionally, it achieves an efficient convergence rate of $\\mathcal{O}(\\frac{1}{T})$ in subsequent learning, where $T$ is the total communication rounds. This rate matches state-of-the-art results. Experimental results show that compared with the Retrain method, PDUDT saves more than 99\\% of unlearning time while achieving comparable unlearning performance."
},
{
"venue": "ICML",
"title": "Rethinking Score Distilling Sampling for 3D Editing and Generation",
"authors": [
"Xingyu Miao",
"Haoran Duan",
"Yang Long",
"Jungong Han"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=1dZgzGTZEO",
"source": "openreview",
"doi": "",
"abstract": "Score Distillation Sampling (SDS) has emerged as a prominent method for text-to-3D generation by leveraging the strengths of 2D diffusion models. However, SDS is limited to generation tasks and lacks the capability to edit existing 3D assets. Conversely, variants of SDS that introduce editing capabilities often can not generate new 3D assets effectively. In this work, we observe that the processes of generation and editing within SDS and its variants have unified underlying gradient terms. Building on this insight, we propose Unified Distillation Sampling (UDS), a method that seamlessly integrates both the generation and editing of 3D assets. Essentially, UDS refines the gradient terms used in vanilla SDS methods, unifying them to support both tasks. Extensive experiments demonstrate that UDS not only outperforms baseline methods in generating 3D assets with richer details but also excels in editing tasks, thereby bridging the gap between 3D generation and editing."
},
{
"venue": "ICML",
"title": "Latent Action Learning Requires Supervision in the Presence of Distractors",
"authors": [
"Alexander Nikulin",
"Ilya Zisman",
"Denis Tarasov",
"Lyubaykin Nikita",
"Andrei Polubarov",
"Igor Kiselev",
"Vladislav Kurenkov"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=2gcEQCT7QW",
"source": "openreview",
"doi": "",
"abstract": "Recently, latent action learning, pioneered by Latent Action Policies (LAPO), have shown remarkable pre-training efficiency on observation-only data, offering potential for leveraging vast amounts of video available on the web for embodied AI. However, prior work has focused on distractor-free data, where changes between observations are primarily explained by ground-truth actions. Unfortunately, real-world videos contain action-correlated distractors that may hinder latent action learning. Using Distracting Control Suite (DCS) we empirically investigate the effect of distractors on latent action learning and demonstrate that LAPO struggle in such scenario. We propose LAOM, a simple LAPO modification that improves the quality of latent actions by **8x**, as measured by linear probing. Importantly, we show that providing supervision with ground-truth actions, as few as 2.5% of the full dataset, during latent action learning improves downstream performance by **4.2x** on average. Our findings suggest that integrating supervision during Latent Action Models (LAM) training is critical in the presence of distractors, challenging the conventional pipeline of first learning LAM and only then decoding from latent to ground-truth actions."
},
{
"venue": "ICML",
"title": "SafeAuto: Knowledge-Enhanced Safe Autonomous Driving with Multimodal Foundation Models",
"authors": [
"Jiawei Zhang",
"Xuan Yang",
"Taiqi Wang",
"Yu Yao",
"Aleksandr Petiushko",
"Bo Li"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=nKJGjovmZz",
"source": "openreview",
"doi": "",
"abstract": "Traditional autonomous driving systems often struggle to connect high-level reasoning with low-level control, leading to suboptimal and sometimes unsafe behaviors. Recent advances in multimodal large language models (MLLMs), which process both visual and textual data, offer an opportunity to unify perception and reasoning. However, effectively embedding precise safety knowledge into MLLMs for autonomous driving remains a significant challenge.\nTo address this, we propose SafeAuto, a framework that enhances MLLM-based autonomous driving by incorporating both unstructured and structured knowledge. First, we introduce a Position-Dependent Cross-Entropy (PDCE) loss to improve low-level control signal predictions when values are represented as text. Second, to explicitly integrate safety knowledge, we develop a reasoning component that translates traffic rules into first-order logic (e.g., \"red light => stop\") and embeds them into a probabilistic graphical model (e.g., Markov Logic Network) to verify predicted actions using recognized environmental attributes.\nAdditionally, our Multimodal Retrieval-Augmented Generation (RAG) model leverages video, control signals, and environmental attributes to learn from past driving experiences. Integrating PDCE, MLN, and Multimodal RAG, SafeAuto outperforms existing baselines across multiple datasets, enabling more accurate, reliable, and safer autonomous driving. The code is available at https://github.com/AI-secure/SafeAuto."
},
{
"venue": "ICML",
"title": "Diverse Prototypical Ensembles Improve Robustness to Subpopulation Shift",
"authors": [
"Nguyen Nhat Minh To",
"Paul F R Wilson",
"Viet Nguyen",
"Mohamed Harmanani",
"Michael Cooper",
"Fahimeh Fooladgar",
"Purang Abolmaesumi",
"Parvin Mousavi",
"Rahul Krishnan"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=qUTiOeM57J",
"source": "openreview",
"doi": "",
"abstract": "Subpopulation shift, characterized by a disparity in subpopulation distribution between the training and target datasets, can significantly degrade the performance of machine learning models. Current solutions to subpopulation shift involve modifying empirical risk minimization with re-weighting strategies to improve generalization. This strategy relies on assumptions about the number and nature of subpopulations and annotations on group membership, which are unavailable for many real-world datasets. Instead, we propose using an ensemble of diverse classifiers to adaptively capture risk associated with subpopulations. Given a feature extractor network, we replace its standard linear classification layer with a mixture of prototypical classifiers, where each member is trained to classify the data while focusing on different features and samples from other members. In empirical evaluation on nine real-world datasets, covering diverse domains and kinds of subpopulation shift, our method of Diverse Prototypical Ensembles (DPEs) often outperforms the prior state-of-the-art in worst-group accuracy. The code is available at https://github.com/minhto2802/dpe4subpop."
},
{
"venue": "ICML",
"title": "Privacy-Shielded Image Compression: Defending Against Exploitation from Vision-Language Pretrained Models",
"authors": [
"Xuelin Shen",
"Jiayin Xu",
"Kangsheng Yin",
"Wenhan Yang"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=olzs3zVsE7",
"source": "openreview",
"doi": "",
"abstract": "The improved semantic understanding of vision-language pretrained (VLP) models has made it increasingly difficult to protect publicly posted images from being exploited by search engines and other similar tools. In this context, this paper seeks to protect users' privacy by implementing defenses at the image compression stage to prevent exploitation. Specifically, we propose a flexible coding method, termed Privacy-Shielded Image Compression (PSIC), that can produce bitstreams with multiple decoding options. By default, the bitstream is decoded to preserve satisfactory perceptual quality while preventing interpretation by VLP models. Our method also retains the original image compression functionality. With a customizable input condition, the proposed scheme can reconstruct the image that preserves its full semantic information. A Conditional Latent Trigger Generation (CLTG) module is proposed to produce bias information based on customizable conditions to guide the decoding process into different reconstructed versions, and an Uncertainty-Aware Encryption-Oriented (UAEO) optimization function is designed to leverage the soft labels inferred from the target VLP model's uncertainty on the training data. This paper further incorporates an adaptive multi-objective optimization strategy to obtain improved encrypting performance and perceptual quality simultaneously within a unified training process. The proposed scheme is plug-and-play and can be seamlessly integrated into most existing Learned Image Compression (LIC) models. Extensive experiments across multiple downstream tasks have demonstrated the effectiveness of our design."
},
{
"venue": "ICML",
"title": "HYGMA: Hypergraph Coordination Networks with Dynamic Grouping for Multi-Agent Reinforcement Learning",
"authors": [
"Chiqiang Liu",
"Dazi Li"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=mgJkeqc685",
"source": "openreview",
"doi": "",
"abstract": "Cooperative multi-agent reinforcement learning faces significant challenges in effectively organizing agent relationships and facilitating information exchange, particularly when agents need to adapt their coordination patterns dynamically. This paper presents a novel framework that integrates dynamic spectral clustering with hypergraph neural networks to enable adaptive group formation and efficient information processing in multi-agent systems. The proposed framework dynamically constructs and updates hypergraph structures through spectral clustering on agents' state histories, enabling higher-order relationships to emerge naturally from agent interactions. The hypergraph structure is enhanced with attention mechanisms for selective information processing, providing an expressive and efficient way to model complex agent relationships. This architecture can be implemented in both value-based and policy-based paradigms through a unified objective combining task performance with structural regularization. Extensive experiments on challenging cooperative tasks demonstrate that our method significantly outperforms state-of-the-art approaches in both sample efficiency and final performance. The code is available at: https://github.com/mysteryelder/HYGMA."
},
{
"venue": "ICML",
"title": "Curvature Enhanced Data Augmentation for Regression",
"authors": [
"Ilya Kaufman",
"Omri Azencot"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=l1sx5KiM7Z",
"source": "openreview",
"doi": "",
"abstract": "Deep learning models with a large number of parameters, often referred to as over-parameterized models, have achieved exceptional performance across various tasks. Despite concerns about overfitting, these models frequently generalize well to unseen data, thanks to effective regularization techniques, with data augmentation being among the most widely used. While data augmentation has shown great success in classification tasks using label-preserving transformations, its application in regression problems has received less attention. Recently, a novel manifold learning approach for generating synthetic data was proposed, utilizing a first-order approximation of the data manifold. Building on this foundation, we present a theoretical framework and practical tools for approximating and sampling general data manifolds. Furthermore, we introduce the Curvature-Enhanced Manifold Sampling (CEMS) method for regression tasks. CEMS leverages a second-order representation of the data manifold to enable efficient sampling and reconstruction of new data points. Extensive evaluations across multiple datasets and comparisons with state-of-the-art methods demonstrate that CEMS delivers superior performance in both in-distribution and out-of-distribution scenarios, while introducing only minimal computational overhead. Code is available at https://github.com/azencot-group/CEMS."
},
{
"venue": "ICML",
"title": "The Surprising Effectiveness of Test-Time Training for Few-Shot Learning",
"authors": [
"Ekin Akyürek",
"Mehul Damani",
"Adam Zweiger",
"Linlu Qiu",
"Han Guo",
"Jyothish Pari",
"Yoon Kim",
"Jacob Andreas"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=asgBo3FNdg",
"source": "openreview",
"doi": "",
"abstract": "Language models (LMs) have shown impressive performance on tasks within their training distribution, but often struggle with structurally novel tasks even when given a small number of in-context task examples. We investigate the effectiveness of test-time training (TTT)—temporarily updating model parameters during inference using a loss derived from input data—as a mechanism for improving LMs' reasoning and few-shot learning capabilities. On the Abstraction and Reasoning Corpus (ARC), performing TTT with in-context examples yields up to $6\\times$ higher accuracy compared to fine-tuned baselines—reaching $53.0\\%$ on the public validation set with an 8B-parameter LM and $61.9\\%$ when ensembled with program-synthesis methods, matching average human performance. On BIG-Bench Hard (BBH), TTT on in-context examples surpasses standard few-shot prompting in the $10$-shot setting by $7.3$ percentage points ($50.5\\%$ to $57.8\\%$). Our findings highlight the limitations of in-context learning for novel tasks and demonstrate the potential of test-time training to enhance language model adaptability."
},
{
"venue": "ICML",
"title": "TOPLOC: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference",
"authors": [
"Jack Min Ong",
"Matthew Di Ferrante",
"Aaron Pazdera",
"Ryan Garner",
"Sami Jaghouar",
"Manveer Basra",
"Max Ryabinin",
"Johannes Hagemann"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=8PJmKfeDdp",
"source": "openreview",
"doi": "",
"abstract": "Large language models (LLMs) have proven to be very capable, but access to frontier models currently relies on inference providers.\nThis introduces trust challenges: how can we be sure that the provider is using the model configuration they claim?\nWe propose TOPLOC, a novel method for verifiable inference that addresses this problem.\nTOPLOC leverages a compact locality-sensitive hashing mechanism for intermediate activations, which can detect unauthorized modifications to models, prompts, or precision with 100\\% accuracy, achieving no false positives or negatives in our empirical evaluations.\nOur approach is robust across diverse hardware configurations, GPU types, and algebraic reorderings, which allows for validation speeds significantly faster than the original inference.\nBy introducing a polynomial encoding scheme, TOPLOC minimizes the memory overhead of the generated proofs by $1000\\times$, requiring only 258 bytes of storage per 32 new tokens, compared to the 262 KB requirement of storing the token embeddings directly for Llama 3.1-8B-Instruct.\nOur method empowers users to verify LLM inference computations efficiently, fostering greater trust and transparency in open ecosystems and laying a foundation for decentralized, verifiable and trustless AI services."
},
{
"venue": "ICML",
"title": "Exploring Invariance in Images through One-way Wave Equations",
"authors": [
"Yinpeng Chen",
"Dongdong Chen",
"Xiyang Dai",
"Mengchen Liu",
"Yinan Feng",
"Youzuo Lin",
"Lu Yuan",
"Zicheng Liu"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=HdogAuhlD5",
"source": "openreview",
"doi": "",
"abstract": "In this paper, we empirically demonstrate that natural images can be reconstructed with high fidelity from compressed representations using a simple first-order norm-plus-linear autoregressive (FINOLA) process—without relying on explicit positional information. Through systematic analysis, we observe that the learned coefficient matrices ($\\mathbf{A}$ and $\\mathbf{B}$) in FINOLA are typically invertible, and their product, $\\mathbf{AB}^{-1}$, is diagonalizable across training runs. This structure enables a striking interpretation: FINOLA’s latent dynamics resemble a system of one-way wave equations evolving in a compressed latent space. Under this framework, each image corresponds to a unique solution of these equations. This offers a new perspective on image invariance, suggesting that the underlying structure of images may be governed by simple, invariant dynamic laws. Our findings shed light on a novel avenue for understanding and modeling visual data through the lens of latent-space dynamics and wave propagation."
},
{
"venue": "ICML",
"title": "Unifying Knowledge from Diverse Datasets to Enhance Spatial-Temporal Modeling: A Granularity-Adaptive Geographical Embedding Approach",
"authors": [
"Zhigaoyuan Wang",
"Ying Sun",
"Hengshu Zhu"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=uPVynwZxch",
"source": "openreview",
"doi": "",
"abstract": "Spatio-temporal forecasting provides potential for discovering evolutionary patterns in geographical scientific data. However, geographical scientific datasets are often manually collected across studies, resulting in limited time spans and data scales. This hinders existing methods that rely on rich historical data for individual entities. In this paper, we argue that heterogeneous datasets from different studies can provide complementary insights into the same underlying system, helping improve predictions for geographical entities with limited historical data. To this end, we propose a Segment Quadtree Geographical Embedding Framework (SQGEF). SQGEF integrates knowledge from datasets with varied target entities, time spans, and observation variables to learn unified representations for multi-granularity entities—including those absent during training. Specifically, we propose a novel data structure, Segment Quadtree, that flexibly accommodates entities of varying granularities. SQGEF not only captures multi-level interactions from grid data but also extracts nested relationships and human-defined boundaries from diverse entities, enabling a comprehensive understanding of complex geographical structures. \nExperiments on real-world datasets demonstrate that SQGEF effectively represents unseen geographical entities and enhances performance for various models."
},
{
"venue": "ICML",
"title": "DiLQR: Differentiable Iterative Linear Quadratic Regulator via Implicit Differentiation",
"authors": [
"Shuyuan Wang",
"Philip D Loewen",
"Michael Forbes",
"Bhushan Gopaluni",
"Wei Pan"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=m2EfTrbv4o",
"source": "openreview",
"doi": "",
"abstract": "While differentiable control has emerged as a powerful paradigm combining model-free flexibility with model-based efficiency, the iterative Linear Quadratic Regulator (iLQR) remains underexplored as a differentiable component. The scalability of differentiating through extended iterations and horizons poses significant challenges, hindering iLQR from being an effective differentiable controller. This paper introduces DiLQR, a framework that facilitates differentiation through iLQR, allowing it to serve as a trainable and differentiable module, either as or within a neural network. A novel aspect of this framework is the analytical solution that it provides for the gradient of an iLQR controller through implicit differentiation, which ensures a constant backward cost regardless of iteration, while producing an accurate gradient. We evaluate our framework on imitation tasks on famous control benchmarks. Our analytical method demonstrates superior computational performance, achieving up to $\\textbf{128x}$ speedup and a minimum of $\\textbf{21x}$ speedup compared to automatic differentiation. Our method also demonstrates superior learning performance ($\\mathbf{10^6x}$) compared to traditional neural network policies and better model loss with differentiable controllers that lack exact analytical gradients. Furthermore, we integrate our module into a larger network with visual inputs to demonstrate the capacity of our method for high-dimensional, fully end-to-end tasks. Codes can be found on the project homepage~\\url{https://sites.google.com/view/dilqr/}."
},
{
"venue": "ICML",
"title": "The Disparate Benefits of Deep Ensembles",
"authors": [
"Kajetan Schweighofer",
"Adrian Arnaiz-Rodriguez",
"Sepp Hochreiter",
"Nuria M Oliver"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=tjPxZiqeHB",
"source": "openreview",
"doi": "",
"abstract": "Ensembles of Deep Neural Networks, Deep Ensembles, are widely used as a simple way to boost predictive performance. However, their impact on algorithmic fairness is not well understood yet. Algorithmic fairness examines how a model's performance varies across socially relevant groups defined by protected attributes such as age, gender, or race. In this work, we explore the interplay between the performance gains from Deep Ensembles and fairness. Our analysis reveals that they unevenly favor different groups, a phenomenon that we term the disparate benefits effect. We empirically investigate this effect using popular facial analysis and medical imaging datasets with protected group attributes and find that it affects multiple established group fairness metrics, including statistical parity and equal opportunity. Furthermore, we identify that the per-group differences in predictive diversity of ensemble members can explain this effect. Finally, we demonstrate that the classical Hardt post-processing method is particularly effective at mitigating the disparate benefits effect of Deep Ensembles by leveraging their better-calibrated predictive distributions."
},
{
"venue": "ICML",
"title": "Power Mean Estimation in Stochastic Continuous Monte-Carlo Tree Search",
"authors": [
"Tuan Quang Dam"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=LL8R2QUEvB",
"source": "openreview",
"doi": "",
"abstract": "Monte Carlo Tree Search (MCTS) has demonstrated success in online planning for deterministic environments, yet significant challenges remain in adapting it to stochastic Markov Decision Processes (MDPs), particularly in continuous state-action spaces. Existing methods, such as HOOT, which combines MCTS with the Hierarchical Optimistic Optimization (HOO) bandit strategy, address continuous spaces but rely on a logarithmic exploration bonus that lacks theoretical guarantees in non-stationary, stochastic settings. Recent advancements, such as POLY-HOOT, introduced a polynomial bonus term to achieve convergence in deterministic MDPs, though a similar theory for stochastic MDPs remains undeveloped.\nIn this paper, we propose a novel MCTS algorithm, Stochastic-Power-HOOT, designed for continuous, stochastic MDPs. Stochastic-Power-HOOT integrates a power mean as a value backup operator, alongside a polynomial exploration bonus to address the non-stationarity inherent in continuous action spaces. Our theoretical analysis establishes that Stochastic-Power-HOOT converges at a polynomial rate of $\\mathcal{O}(n^{-\\zeta})$, $\\zeta \\in (0,1/2)$, where \\( n \\) is the number of visited trajectories, thereby extending the non-asymptotic convergence guarantees of POLY-HOOT to stochastic environments. Experimental results on stochastic tasks validate our theoretical findings, demonstrating the effectiveness of Stochastic-Power-HOOT in continuous, stochastic domains."
},
{
"venue": "ICML",
"title": "Provable In-Context Vector Arithmetic via Retrieving Task Concepts",
"authors": [
"Dake Bu",
"Wei Huang",
"Andi Han",
"Atsushi Nitanda",
"Qingfu Zhang",
"Hau-San Wong",
"Taiji Suzuki"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=DbUmeNnNpt",
"source": "openreview",
"doi": "",
"abstract": "In-context learning (ICL) has garnered significant attention for its ability to grasp functions/tasks from demonstrations. Recent studies suggest the presence of a latent **task/function vector** in LLMs during ICL. Merullo et al. (2024) showed that LLMs leverage this vector alongside the residual stream for Word2Vec-like vector arithmetic, solving factual-recall ICL tasks. Additionally, recent work empirically highlighted the key role of Question-Answer data in enhancing factual-recall capabilities. Despite these insights, a theoretical explanation remains elusive. To move one step forward, we propose a theoretical framework building on empirically grounded *hierarchical* concept modeling. We develop an optimization theory, showing how nonlinear residual transformers trained via gradient descent on cross-entropy loss perform factual-recall ICL tasks via vector arithmetic. We prove 0-1 loss convergence and show the strong generalization, including robustness to concept recombination and distribution shifts. These results elucidate the advantages of transformers over static embedding predecessors. Empirical simulations corroborate our theoretical insights."
},
{
"venue": "ICML",
"title": "Contextual Bandits for Unbounded Context Distributions",
"authors": [
"Puning Zhao",
"Rongfei Fan",
"Shaowei Wang",
"Li Shen",
"Qixin Zhang",
"ZongKe",
"Tianhang Zheng"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=gGY9TNVYs3",
"source": "openreview",
"doi": "",
"abstract": "Nonparametric contextual bandit is an important model of sequential decision making problems. Under $\\alpha$-Tsybakov margin condition, existing research has established a regret bound of $\\tilde{O}\\left(T^{1-\\frac{\\alpha+1}{d+2}}\\right)$ for bounded supports. However, the optimal regret with unbounded contexts has not been analyzed. The challenge of solving contextual bandit problems with unbounded support is to achieve both exploration-exploitation tradeoff and bias-variance tradeoff simultaneously. In this paper, we solve the nonparametric contextual bandit problem with unbounded contexts. We propose two nearest neighbor methods combined with UCB exploration. The first method uses a fixed $k$. Our analysis shows that this method achieves minimax optimal regret under a weak margin condition and relatively light-tailed context distributions. The second method uses adaptive $k$. By a proper data-driven selection of $k$, this method achieves an expected regret of $\\tilde{O}\\left(T^{1-\\frac{(\\alpha+1)\\beta}{\\alpha+(d+2)\\beta}}+T^{1-\\beta}\\right)$, in which $\\beta$ is a parameter describing the tail strength. This bound matches the minimax lower bound up to logarithm factors, indicating that the second method is approximately optimal."
},
{
"venue": "ICML",
"title": "PiD: Generalized AI-Generated Images Detection with Pixelwise Decomposition Residuals",
"authors": [
"Xinghe Fu",
"Zhiyuan Yan",
"Zheng Yang",
"Taiping Yao",
"Yandan Zhao",
"Shouhong Ding",
"Xi Li"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=gye2zYytx6",
"source": "openreview",
"doi": "",
"abstract": "Fake images, created by recently advanced generative models, have become increasingly indistinguishable from real ones, making their detection crucial, urgent, and challenging. This paper introduces PiD (Pixelwise Decomposition Residuals), a novel detection method that focuses on residual signals within images. Generative models are designed to optimize high-level semantic content (principal components), often overlooking low-level signals (residual components). PiD leverages this observation by disentangling residual components from images, encouraging the model to uncover more underlying and general forgery clues independent of semantic content. Compared to prior approaches that rely on reconstruction techniques or high-frequency information, PiD is computationally efficient and does not rely on any generative models for reconstruction. Specifically, PiD operates at the pixel level, mapping the pixel vector to another color space (e.g., YUV) and then quantizing the vector. The pixel vector is mapped back to the RGB space and the quantization loss is taken as the residual for AIGC detection. Our experiment results are striking and highly surprising: PiD achieves 98% accuracy on the widely used GenImage benchmark, highlighting the effectiveness and generalization performance."
},
{
"venue": "ICML",
"title": "CAT: Contrastive Adversarial Training for Evaluating the Robustness of Protective Perturbations in Latent Diffusion Models",
"authors": [
"Sen Peng",
"Mingyue Wang",
"Jianfei He",
"Jijia Yang",
"Xiaohua Jia"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=5of0l7eUau",
"source": "openreview",
"doi": "",
"abstract": "Latent diffusion models have recently demonstrated superior capabilities in many downstream image synthesis tasks. \nHowever, customization of latent diffusion models using unauthorized data can severely compromise the privacy and intellectual property rights of data owners.\nAdversarial examples as protective perturbations have been developed to defend against unauthorized data usage by introducing imperceptible noise to customization samples, preventing diffusion models from effectively learning them.\nIn this paper, we first reveal that the primary reason adversarial examples are effective as protective perturbations in latent diffusion models is the distortion of their latent representations, as demonstrated through qualitative and quantitative experiments.\nWe then propose the Contrastive Adversarial Training (CAT) utilizing lightweight adapters as an adaptive attack against these protection methods, highlighting their lack of robustness. \nExtensive experiments demonstrate that our CAT method significantly reduces the effectiveness of protective perturbations in customization, urging the community to reconsider and improve the robustness of existing protective perturbations. \nThe code is available at \\url{https://github.com/senp98/CAT}."
},
{
"venue": "ICML",
"title": "Learning Condensed Graph via Differentiable Atom Mapping for Reaction Yield Prediction",
"authors": [
"Ankit Ghosh",
"Gargee Kashyap",
"Sarthak Mittal",
"Nupur Jain",
"Raghavan B Sunoj",
"Abir De"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=sqjQ6p56GR",
"source": "openreview",
"doi": "",
"abstract": "Yield of chemical reactions generally depends on the activation barrier, i.e., the energy difference between the reactant and the transition state. Computing the transition state from the reactant and product graphs requires prior knowledge of the correct node alignment (i.e., atom mapping), which is not available in yield prediction datasets. In this work, we propose YieldNet, a neural yield prediction model, which tackles these challenges. Here, we first approximate the atom mapping between the reactants and products using a differentiable node alignment network. We then use this approximate atom mapping to obtain a noisy realization of the condensed graph of reaction (CGR), which is a supergraph encompassing both the reactants and products. This CGR serves as a surrogate for the transition state graph structure. The CGR embeddings of different steps in a multi-step reaction are then passed into a transformer-guided reaction path encoder.\nOur experiments show that YieldNet can predict the yield more accurately than the baselines. Furthermore, the model is trained only under the distant supervision of yield values, without requiring fine-grained supervision of atom mapping."
},
{
"venue": "ICML",
"title": "What Limits Bidirectional Model's Generative Capabilities? A Uni-Bi-Directional Mixture-of-Expert Method For Bidirectional Fine-tuning",
"authors": [
"Zuchao Li",
"Yonghua Hei",
"Qiwei Li",
"Lefei Zhang",
"Ping Wang",
"hai zhao",
"Baoyuan Qi",
"Liu Guoming"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=kPqvx2mvec",
"source": "openreview",
"doi": "",
"abstract": "Large Language Models (LLMs) excel in generation tasks, yet their causal attention mechanisms limit performance in embedding tasks. While bidirectional modeling may enhance embeddings, naively fine-tuning unidirectional models bidirectionally severely degrades generative performance.\nTo investigate this trade-off, we analyze attention weights as dependence indicators and find that bidirectional fine-tuning increases subsequent dependence, impairing unidirectional generation. Through systematic Transformer module evaluations, we discover the FFN layer is least affected by such dependence. Leveraging this discovery, we propose UBMoE-LLM, a novel Uni-Bi-directional Mixture-of-Experts LLM, which integrates the original unidirectional FFN with a bidirectionally fine-tuned FFN via unsupervised contrastive learning. This MoE-based approach enhances embedding performance while preserving robust generation.\nExtensive experiments across diverse datasets and model scales validate our attention dependence metric and demonstrate UBMoE-LLM’s superior generative quality and reduced hallucination. Code is available at: https://github.com/heiyonghua/ubmoe_llm."
},
{
"venue": "ICML",
"title": "GTR: A General, Multi-View, and Dynamic Framework for Trajectory Representation Learning",
"authors": [
"Xiangheng Wang",
"Ziquan Fang",
"Chenglong Huang",
"Danlei Hu",
"Lu Chen",
"Yunjun Gao"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=ehcWKZ2nEn",
"source": "openreview",
"doi": "",
"abstract": "Trajectory representation learning aims to transform raw trajectory data into compact and low-dimensional vectors that are suitable for downstream analysis. However, most existing methods adopt either a free-space view or a road-network view during the learning process, which limits their ability to capture the complex, multi-view spatiotemporal features inherent in trajectory data. Moreover, these approaches rely on task-specific model training, restricting their generalizability and effectiveness for diverse analysis tasks. To this end, we propose GTR, a general, multi-view, and dynamic Trajectory Representation framework built on a pre-train and fine-tune architecture. Specifically, GTR introduces a multi-view encoder that captures the intrinsic multi-view spatiotemporal features. Based on the pre-train and fine-tune architecture, we provide the spatio-temporal fusion pre-training with a spatio-temporal mixture of experts to dynamically combine spatial and temporal features, enabling seamless adaptation to diverse trajectory analysis tasks. Furthermore, we propose an online frozen-hot updating strategy to efficiently update the representation model, accommodating the dynamic nature of trajectory data. Extensive experiments on two real-world datasets demonstrate that GTR consistently outperforms 15 state-of-the-art methods across 6 mainstream trajectory analysis tasks. All source code and data are available at https://github.com/ZJU-DAILY/GTR."
},
{
"venue": "ICML",
"title": "µnit Scaling: Simple and Scalable FP8 LLM Training",
"authors": [
"Saaketh Narayan",
"Abhay Gupta",
"Mansheej Paul",
"Davis Blalock"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=qOLjAhxZgm",
"source": "openreview",
"doi": "",
"abstract": "Large language model training with 8-bit floating point (FP8) formats promises significant efficiency improvements, but reduced numerical precision makes training challenging. It is currently possible to train in FP8 only if one is willing to tune various hyperparameters, reduce model scale, or accept the overhead of computing dynamic scale factors. We demonstrate simple, scalable FP8 training that requires no dynamic scaling factors or special hyperparameters, even at large model sizes. Our method, \\textit{µnit Scaling (µS)}, also enables simple hyperparameter transfer across model widths, matched numerics across training and inference, and other desirable properties. µnit Scaling is straightforward to implement, consisting of a set of minimal interventions based on a first-principles analysis of transformer operations. We validate our method by training models with parameters ranging from 1B to 13B, performing all hidden linear layer computations in FP8. We achieve quality equal to higher-precision baselines while also training up to 33% faster."
},
{
"venue": "ICML",
"title": "Contextual Optimization Under Model Misspecification: A Tractable and Generalizable Approach",
"authors": [
"Omar Bennouna",
"Jiawei Zhang",
"Saurabh Amin",
"Asuman E. Ozdaglar"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=e3NNvqD7wA",
"source": "openreview",
"doi": "",
"abstract": "Contextual optimization problems are prevalent in decision-making applications where historical data and contextual features are used to learn predictive models that inform optimal actions. However, practical applications often suffer from model misspecification due to incomplete knowledge of the underlying data-generating process, leading to suboptimal decisions. Existing approaches primarily address the well-specified case, leaving a critical gap in handling misspecified models. In this paper, we propose a novel Integrated Learning and Optimization (ILO) framework that explicitly accounts for model misspecification by introducing a tractable surrogate loss function with strong theoretical guarantees on generalizability, tractability, and optimality. Our surrogate loss aligns with the true decision performance objective, ensuring robustness to misspecification without imposing restrictive assumptions. The proposed approach effectively mitigates the challenges of non-convexity and non-smoothness in the target loss function, leading to efficient optimization procedures. We provide rigorous theoretical analysis and experimental validation, demonstrating superior performance compared to state-of-the-art methods. Our work offers a principled solution to the practically relevant challenge of model misspecification in contextual optimization."
},
{
"venue": "ICML",
"title": "NextCoder: Robust Adaptation of Code LMs to Diverse Code Edits",
"authors": [
"Tushar Aggarwal",
"Swayam Singh",
"Abhijeet Awasthi",
"Aditya Kanade",
"Nagarajan Natarajan"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=3B6fF1PxYD",
"source": "openreview",
"doi": "",
"abstract": "Software engineering activities frequently involve edits to existing code. However, contemporary code language models (LMs) lack the ability to handle diverse types of code-edit requirements. In this work, we attempt to overcome this shortcoming through (1) a novel synthetic data generation pipeline and (2) a robust model adaptation algorithm. Starting with seed code examples and diverse editing criteria, our pipeline generates high-quality samples comprising original and modified code, along with natural language instructions in different styles and verbosity. Today's code LMs come bundled with strong abilities, such as code generation and instruction following, which should not be lost due to fine-tuning. To ensure this, we propose a novel adaptation algorithm, SeleKT, that (a) leverages a dense gradient-based step to identify the weights that are most important for code editing, and (b) does a sparse projection onto the base model to avoid overfitting. Using our approach, we obtain a new series of models NextCoder (adapted from QwenCoder-2.5) that achieves strong results on five code-editing benchmarks, outperforming comparable size models and even several larger ones. We show the generality of our approach on two model families DeepSeekCoder and QwenCoder), compare against other fine-tuning approaches, and demonstrate robustness by showing retention of code generation and general problem-solving abilities post adaptation. We opensource the models, synthetic dataset, and implementation at http://aka.ms/nextcoder."
},
{
"venue": "ICML",
"title": "RoSTE: An Efficient Quantization-Aware Supervised Fine-Tuning Approach for Large Language Models",
"authors": [
"Quan Wei",
"Chung-Yiu Yau",
"Hoi To Wai",
"Yang Zhao",
"Dongyeop Kang",
"Youngsuk Park",
"Mingyi Hong"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=h30EzoI3s0",
"source": "openreview",
"doi": "",
"abstract": "Supervised fine-tuning is a standard method for adapting pre-trained large language models (LLMs) to downstream tasks. Quantization has been recently studied as a post-training technique for efficient LLM deployment. To obtain quantized fine-tuned LLMs, conventional pipelines would first fine-tune the pre-trained models, followed by post-training quantization. This often yields suboptimal performance as it fails to leverage the synergy between fine-tuning and quantization. To effectively realize low-bit quantization of weights, activations and KV caches in LLMs, we propose an algorithm named Rotated Straight-Through-Estimator (RoSTE), which combines quantization-aware supervised fine-tuning (QA-SFT) with an adaptive rotation strategy that identifies an effective rotation configuration to reduce activation outliers. We provide theoretical insights on RoSTE by analyzing its prediction error when applied to an overparameterized least square quantized training problem. Our findings reveal that the prediction error is directly proportional to the quantization error of the converged weights, which can be effectively managed through an optimized rotation configuration. Experiments on Pythia, Qwen and Llama models of different sizes demonstrate the effectiveness of RoSTE. Compared to existing post-SFT quantization baselines, our method consistently achieves superior performances across various tasks and different LLM architectures. Our code is available at https://github.com/OptimAI-Lab/RoSTE."
},
{
"venue": "ICML",
"title": "Reflection-Bench: Evaluating Epistemic Agency in Large Language Models",
"authors": [
"Lingyu Li",
"Yixu Wang",
"Haiquan Zhao",
"Shuqi Kong",
"Yan Teng",
"Chunbo Li",
"Yingchun Wang"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=eff38SdyvN",
"source": "openreview",
"doi": "",
"abstract": "With large language models (LLMs) increasingly deployed as cognitive engines for AI agents, the reliability and effectiveness critically hinge on their intrinsic epistemic agency, which remains understudied. Epistemic agency, the ability to flexibly construct, adapt, and monitor beliefs about dynamic environments, represents a base-model-level capacity independent of specific tools, modules, or applications. We characterize the holistic process underlying epistemic agency, which unfolds in seven interrelated dimensions: prediction, decision-making, perception, memory, counterfactual thinking, belief updating, and meta-reflection. Correspondingly, we propose Reflection-Bench, a cognitive-psychology-inspired benchmark consisting of seven tasks with long-term relevance and minimization of data leakage. Through a comprehensive evaluation of 16 models using three prompting strategies, we identify a clear three-tier performance hierarchy and significant limitations of current LLMs, particularly in meta-reflection capabilities. While state-of-the-art LLMs demonstrate rudimentary signs of epistemic agency, our findings suggest several promising research directions, including enhancing core cognitive functions, improving cross-functional coordination, and developing adaptive processing mechanisms. Our code and data are available at https://github.com/AI45Lab/ReflectionBench."
},
{
"venue": "ICML",
"title": "COGNATE: Acceleration of Sparse Tensor Programs on Emerging Hardware using Transfer Learning",
"authors": [
"Chamika Sudusinghe",
"Gerasimos Gerogiannis",
"Damitha Lenadora",
"Charles Block",
"Josep Torrellas",
"Charith Mendis"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=EV0itGFjmm",
"source": "openreview",
"doi": "",
"abstract": "Sparse tensor programs are essential in deep learning and graph analytics, driving the need for optimized processing. To meet this demand, specialized hardware accelerators are being developed. Optimizing these programs for accelerators is challenging for two reasons: program performance is highly sensitive to variations in sparse inputs, and early-stage accelerators rely on expensive simulators. Therefore, ML-based cost models used for optimizing such programs on general-purpose hardware are often ineffective for early-stage accelerators, as they require large datasets for proper training. To this end, we introduce COGNATE, a novel framework that leverages inexpensive data samples from general-purpose hardware (e.g., CPUs) to train cost models, followed by few-shot fine-tuning on emerging hardware. COGNATE exploits the homogeneity of input features across hardware platforms while effectively mitigating heterogeneity, enabling cost model training with just 5% of the data samples needed by accelerator-specific models to achieve comparable performance. We conduct extensive experiments to demonstrate that COGNATE outperforms existing techniques, achieving average speedups of 1.47× (up to 5.46×) for SpMM and 1.39× (up to 4.22×) for SDDMM."
},
{
"venue": "ICML",
"title": "Maximal Update Parametrization and Zero-Shot Hyperparameter Transfer for Fourier Neural Operators",
"authors": [
"Shanda Li",
"Shinjae Yoo",
"Yiming Yang"
],
"year": 2025,
"pdf_url": "https://openreview.net/pdf?id=fHt4Nau7FW",
"source": "openreview",
"doi": "",
"abstract": "Fourier Neural Operators (FNOs) offer a principled approach for solving complex partial differential equations (PDEs). However, scaling them to handle more complex PDEs requires increasing the number of Fourier modes, which significantly expands the number of model parameters and makes hyperparameter tuning computationally impractical. To address this, we introduce $\\mu$**Transfer-FNO**, a zero-shot hyperparameter transfer technique that enables optimal configurations, tuned on smaller FNOs, to be directly applied to billion-parameter FNOs _without_ additional tuning. Building on the Maximal Update Parametrization ($\\mu$P) framework, we mathematically derive a parametrization scheme that facilitates the transfer of optimal hyperparameters across models with different numbers of Fourier modes in FNOs, which is validated through extensive experiments on various PDEs. Our empirical study shows that $\\mu$Transfer-FNO reduces computational cost for tuning hyperparameters on large FNOs while maintaining or improving accuracy."
}
]