Upload hard_negatives.json with huggingface_hub

hard_negatives.json

@@ -657,16 +657,6 @@
       "contaminated": false,
       "high_similarity": true
     },
-    {
-      "paper_id": "6eca224e8eef0ae8d34f66e5380476ca39420812",
-      "arxiv_id": null,
-      "title": "Retrieval-Augmented Large Language Models for Robust Context-Aware Natural Language Understanding",
-      "abstract": "Large Language Models (LLMs) have been shown to have remarkable capabilities in natural language\nunderstanding; however, they still have some limitations such as the outdated knowledge, the lack of domain-specific\nawareness and the hallucination of incorrect information. These problems are induced by the fact that LLMs are mainly\nbased on parametric knowledge stored during the training process, that is not dynamically updated and verified . To combat\nsuch challenges, this paper introduces an improved Retrieval-Augmented Generation (RAG) to address these underlying\nchallenges which combines an improved context aware retrieval mechanism with the gating based prompt augmentation\nstrategy. The proposed approach selectively filters and ranks the retrieved documents based on context-awareness gate\nbefore injecting them to the LLM, which would improve the relevance and reduce the noise in the generated responses.In\nthe paper we validate the proposed method using benchmark data such as SQuAD, domain-specific question answering data\nsets as well as dialogue data sets where we compare with baseline models such as vanilla LLMs and standard RAG pipelines.\nExperimental results show that our method can provide much better results in terms of Exact Match (EM), F1-score and\nFact consistency compared to traditional methods. These findings are consistent with recent studies showing the value of\nRAG in enhancing factual grounding and reducing hallucinations in LLMs 1.\n\uf0d8 Contributions:\nIn this paper, we propose a novel context-aware RAG architecture, which provides a retrieval filtering mechanism.\nFollowing the review, we design an improved prompt integration strategy for improved knowledge grounding. We\nempirically show better performance on several NLP benchmarks.",
-      "year": 2026,
-      "score": 0.8421,
-      "source": "keyword_search+arxiv_recs",
-      "contaminated": false
-    },
     {
       "paper_id": "3becf3cee2ec9db3b4bdae0c4504cf6821a6d8e6",
       "arxiv_id": null,
@@ -707,6 +697,17 @@
       "source": "manual_web_r3",
       "source_url": "https://arxiv.org/pdf/2506.00054",
       "contaminated": false
+    },
+    {
+      "paper_id": "web_b71ed17dbba271ad",
+      "arxiv_id": null,
+      "title": "A Survey on RAG Meeting LLMs: Towards Retrieval-Augmented",
+      "abstract": "A Survey on RAG Meeting LLMs: Towards Retrieval-Augmented Large Language Models Wenqi Fan                                             Yujuan Ding\u2217                                 Liangbo Ning wenqifan03@gmail.com                                       dingyujuan385@gmail.com                     BigLemon1123@gmail.com The Hong Kong Polytechnic                                   The Hong Kong Polytechnic                   The Hong Kong Polytechnic University, HK SAR                                          University, HK SAR                          University, HK SAR Shijie Wang                                             Hengyun Li                                     Dawei Yin shijie.wang@connect.polyu.hk                                 neilhengyun.li@polyu.edu.hk                      yindawei@acm.org arXiv:2405.06211v3 [cs.CL] 17 Jun 2024 The Hong Kong Polytechnic                                   The Hong Kong Polytechnic                         Baidu Inc, China University, HK SAR                                          University, HK SAR Tat-Seng Chua                                         Qing Li dcscts@nus.edu.sg                                 csqli@comp.polyu.edu.hk National University of Singapore,                      The Hong Kong Polytechnic Singapore                                       University, HK SAR ABSTRACT                                                                               1   INTRODUCTION As one of the most advanced techniques in AI, Retrieval-",
+      "year": null,
+      "score": 0.4566,
+      "source": "manual_web_r3",
+      "source_url": "https://arxiv.org/pdf/2405.06211",
+      "contaminated": false
     }
   ],
   "honor_watch_gs_pro.pdf": [
@@ -2259,17 +2260,6 @@
       "source_url": "https://internetsociety.org/wp-content/uploads/2017/08/Mobile20Internet20Usage20Trends20in20Asia-Pacific.pdf",
       "contaminated": false
     },
-    {
-      "paper_id": "web_97449246e38ae35d",
-      "arxiv_id": null,
-      "title": "[Skip to content](#wrapper)",
-      "abstract": "[Skip to content](#wrapper) [![GSMA logo](https://www.gsma.com/wp-content/themes/theme_gsma_2023/images_2023/GSMA-Logo-White-RGB.png)](https://www.gsma.com/) * [Solutions and impact](https://www.gsma.com/solutions-and-impact/) Solutions and impact [Solutions and impact](https://www.gsma.com/solutions-and-impact/) - What we do to help business and society thrive + [Industry Services](https://www.gsmaservices.com/) Industry Services - [Device Services](https://www.gsmaservices.com/device-services/) - [Network Services](https://www.gsmaservices.com/network-services/) - [Assurance Services](https://www.gsmaservices.com/assurance-services/) - [Industry Services](https://www.gsmaservices.com/) + [Training and consulting](https://www.gsma.com/solutions-and-impact/training-and-consulting/) Training and consulting - [GSMA Advance](https://www.gsmaadvance.com/) - [GSMA Capacity Building](https://www.gsmatraining.com/) - [GSMA Intelligence](https://www.gsmaintelligence.com/) - [Training and consulting](https://www.gsma.com/solutions-and-impact/training-and-consulting/) + [Connectivity for Good](https://www.gsma.com/solutions-and-impact/connectivity-for-good/) Connectivity for Good - [External Affairs](https://www.gsma.com/solutions-and-impact/connectivity-for-good/external-affairs/) - [Government convening](https://www.gsma.com/solutions-and-impact/connectivity-for-good/government-convening/) - [Mobile for Development](https://www.gsma.com/solutions-and-impact/connectivity-for-good/mobi",
-      "year": null,
-      "score": 0.0,
-      "source": "manual_web_r2",
-      "source_url": "https://www.gsma.com/r/wp-content/uploads/2021/09/The-State-of-Mobile-Internet-Connectivity-Report-2021.pdf",
-      "contaminated": false
-    },
     {
       "paper_id": "web_c4b48f5409e60c35",
       "arxiv_id": null,
@@ -2324,6 +2314,17 @@
       "source": "manual_web_r3",
       "source_url": "https://www.gsma.com/solutions-and-impact/connectivity-for-good/mobile-economy/wp-content/uploads/2024/07/240724-Mobile-Economy-Asia-Pacific-2024-FINAL.pdf",
       "contaminated": false
+    },
+    {
+      "paper_id": "web_a14c73b2e0493281",
+      "arxiv_id": null,
+      "title": "The Mobile",
+      "abstract": "The Mobile Economy China 2025 The GSMA is a global organisation unifying the mobile ecosystem to discover, develop and deliver innovation foundational to positive business environments and societal change. Our vision is to unlock the full power of connectivity so that people, industry and society thrive. Representing mobile operators and organisations across the mobile ecosystem and adjacent industries, the GSMA delivers for its members across three broad pillars: Connectivity for Good, Industry Services and Solutions, and Outreach. This activity includes advancing policy, tackling today\u2019s biggest societal challenges, underpinning the technology and interoperability that make mobile work, and providing the world\u2019s largest platform to convene the mobile ecosystem at the MWC and M360 series of events. We invite you to find out more at gsma.com GSMA Intelligence is the definitive source of global mobile operator data, analysis and forecasts, and publisher of authoritative industry reports and research. Our data covers every operator group, network and MVNO in every country worldwide \u2013 from Afghanistan to Zimbabwe. It is the most accurate and complete set of industry metrics available, comprising tens of millions of individual data points, updated daily. GSMA Intelligence is relied on by leading operators, vendors, regulators, financial institutions and third-party industry players, to support strategic decision-making and long- term investment planning. The data is used as an in",
+      "year": null,
+      "score": 0.2284,
+      "source": "manual_web_r3",
+      "source_url": "https://www.gsma.com/solutions-and-impact/connectivity-for-good/mobile-economy/wp-content/uploads/2025/04/10042025-The-Mobile-Economy-China-2025.pdf",
+      "contaminated": false
     }
   ],
   "2310.07609v1.pdf": [
@@ -2424,16 +2425,6 @@
       "contaminated": false,
       "high_similarity": true
     },
-    {
-      "paper_id": "097116ebde910c77769eb155f32c2dd11663d80e",
-      "arxiv_id": null,
-      "title": "Personalized novelty-aware recommendation in social recommender systems: A Framework",
-      "abstract": "This paper introduces a novel framework for improving social recommender systems by incorporating a personalized notion of item novelty grounded in user\u2019s social interactions. Unlike conventional approaches that treat novelty as a static, item-specific feature, our method estimates the novelty of each item relative to a given user by analyzing behavioral patterns within the user\u2019s social communities. Additionally, we model each user\u2019s individual tendency toward novel content, allowing for personalized calibration of the novelty\u2013relevance trade-off in recommendations. The proposed method operates independently of the underlying recommendation algorithm and can be seamlessly integrated as a post-processing step over candidate lists generated by various base models. Experimental evaluations on two real-world datasets\u2014Epinions, and LastFM\u2014demonstrate that our framework consistently enhances diversity, coverage, and novelty while preserving recommendation relevance. These findings underscore the value of socially contextualized and user-personalized novelty modeling in elevating the effectiveness and user satisfaction of recommender systems.",
-      "year": 2026,
-      "score": 0.7778,
-      "source": "keyword_search+arxiv_recs",
-      "contaminated": false
-    },
     {
       "paper_id": "4136563718c0961bcb43f158b2bc26912772ecb7",
       "arxiv_id": null,
@@ -2497,6 +2488,17 @@
       "source": "manual_web_r3",
       "source_url": "https://arxiv.org/pdf/1812.11422v1",
       "contaminated": false
+    },
+    {
+      "paper_id": "web_289916b20eaef9df",
+      "arxiv_id": null,
+      "title": "IEEE International Conference on Data Mining (ICDM), Workshop on User Modeling and Recommendation (UMRec), Washington, DC, USA, 2025",
+      "abstract": "IEEE International Conference on Data Mining (ICDM), Workshop on User Modeling and Recommendation (UMRec), Washington, DC, USA, 2025 Effectiveness of LLMs in Temporal User Profiling for Recommendation Milad Sabouri\u2217 , Masoud Mansoury\u2020 , Kun Lin\u2217 , Bamshad Mobasher\u2217 \u2217 DePaul University, USA Email: msabouri@depaul.edu, klin13@depaul.edu, mobasher@cs.depaul.edu \u2020 Delft University of Technology, Netherlands Email: m.mansoury@tudelft.nl Abstract\u2014Effectively modeling the dynamic nature of user            17% in Recall@10 and 14% in NDCG@10, benefits are preferences is crucial for enhancing recommendation accuracy           less pronounced in sparser environments like Video Games. arXiv:2511.00176v1 [cs.IR] 31 Oct 2025 and fostering transparency in recommender systems. Traditional         Our analysis of these results leads us to hypothesize that user profiling often overlooks the distinction between transitory short-term interests and stable long-term preferences. This paper      this nuanced capability is particularly evident where short- examines the capability of leveraging Large Language Models            term and long-term preferences are more clearly separable (LLMs) to capture these temporal dynamics, generating richer           (e.g., Movies&TV), versus domains with more stable user user representations through distinct short-term and long-term         profiles (e.g., Video Games). This highlights a critical trade-off textual summaries of interaction histories. Our observat",
+      "year": null,
+      "score": 0.1094,
+      "source": "manual_web_r3",
+      "source_url": "https://arxiv.org/pdf/2511.00176v1",
+      "contaminated": false
     }
   ],
   "2305.13186v3.pdf": [
@@ -2910,17 +2912,6 @@
       "source_url": "https://www.utrgv.edu/graduate/_files/documents/grad-college-international-viewbook.pdf",
       "contaminated": false
     },
-    {
-      "paper_id": "web_8c2aa0d4036d5559",
-      "arxiv_id": null,
-      "title": "* [Skip to content](#main-content)",
-      "abstract": "* [Skip to content](#main-content) * [Skip to search](#header-search) * [Behind the Science magazine](/en-gb/behind-the-science-magazine/ \"Behind the Science magazine\") * [Company](/en-gb/company/ \"Company\") + [Company](/en-gb/company/) + [Purpose, strategy and culture](/en-gb/company/purpose-strategy-and-culture/ \"Purpose, strategy and culture\") + [Outstanding people](/en-gb/company/outstanding-people/ \"Outstanding people\") - [Outstanding people](/en-gb/company/outstanding-people/) - [Great place to work](/en-gb/company/outstanding-people/great-place-to-work/ \"Great place to work\") - [Meet our veterans](/en-gb/company/outstanding-people/meet-our-veterans/ \"Meet our veterans\") - [Working together](/en-gb/company/outstanding-people/working-together/ \"Working together\") - [My GSK story](/en-gb/company/outstanding-people/my-gsk-story/ \"My GSK story\") - [Innovating together](/en-gb/company/outstanding-people/innovating-together/ \"Innovating together\") + [GSK at a glance](/en-gb/company/gsk-at-a-glance/ \"GSK at a glance\") + [History and heritage](/en-gb/company/history-and-heritage/ \"History and heritage\") + [Board of Directors and Executive Committee](/en-gb/company/board-of-directors-and-leadership-team/ \"Board of Directors and Executive Committee\") + [Board information](/en-gb/company/board-information/ \"Board information\") + [Governance](/en-gb/company/governance/ \"Governance\") - [Governance](/en-gb/company/governance/) - [Committees](/en-gb/company/governance/committees/ \"Com",
-      "year": null,
-      "score": 0.0604,
-      "source": "manual_web_r2",
-      "source_url": "https://www.gsk.com/media/10207/university-student-and-graduate-brochure.pdf",
-      "contaminated": false
-    },
     {
       "paper_id": "web_7c476895c4da1bbe",
       "arxiv_id": null,
@@ -2942,6 +2933,17 @@
       "source": "manual_web_r3",
       "source_url": "https://cgsnet.org/wp-content/uploads/2022/01/Grad_Study_In_The_US_FINAL.pdf",
       "contaminated": false
+    },
+    {
+      "paper_id": "web_e92b2ad86c154315",
+      "arxiv_id": null,
+      "title": "EST 1895",
+      "abstract": "EST 1895 years GRADUATE STUDY 2026/27 Contents Welcome                               1 Why LSE?                              5 Research                              7 Study at LSE                         11 Support                              12 Volunteering                         14 Careers                              15 Our campus                           17 Accommodation                        22 Applying to LSE                      24 Fees and funding                     28 Meet, visit and discover LSE 29 Our graduate programmes              32 Equity, diversity and inclusion 46 KEY DATES                            27 May 2026 Second funding application 8 October 2025                       deadline for some research Graduate applications open           programmes 10 December 2025 Early funding application deadline   TERM DATES for some research programmes         Autumn Term January 2026                         Monday 28 September 2026 First decisions issued               Friday 11 December 2026 14 January 2026                      Winter Term First funding application deadline   Monday 11 January 2027 for some research programmes         Thursday 25 March 2027 23 April 2026                        Spring Term Funding application deadline for     Monday 26 April 2027 taught master\u2019s study                Friday 11 June 2027 B Welcome LSE is a special institution. Among the world\u2019s leading universities for the study of the social sciences, it attracts extraordinarily tal",
+      "year": null,
+      "score": 0.1727,
+      "source": "manual_web_r3",
+      "source_url": "https://www.lse.ac.uk/study-at-lse/Assets/PDF/Prospectus-and-Brochures/graduate-guide.pdf",
+      "contaminated": false
     }
   ],
   "afe620b9beac86c1027b96d31d396407.pdf": [
@@ -5768,16 +5770,6 @@
       "source": "keyword_search+arxiv_recs",
       "contaminated": false
     },
-    {
-      "paper_id": "2353a8c93abaf4f2e7f3d412ed9c175207c6fe97",
-      "arxiv_id": null,
-      "title": "Two-Stage Angular Alignment for Positive-Unlabeled Learning",
-      "abstract": ": Positive-Unlabeled (PU) learning addresses the binary classification problem where only positive and unlabeled data are available\u2014a setting common in applications such as medical diagnosis and web mining. We introduce a novel two-stage approach based on angular alignment in feature space, where a learnable prototype vector represents the directional centroid of the positive class. In the first stage, the model aligns labeled positives toward this prototype to promote angular compactness; in the second, it repels overly similar unlabeled instances to refine the decision boundary without prematurely assigning negative labels. Our method employs a directional loss inspired by von Mises\u2013Fisher geometry, a dynamic stage-switching curriculum, and maintains a highly parameter-efficient design. Experiments on CIFAR-10 and SVHN demonstrate strong performance and competitive results compared to state-of-the-art PU learning methods. The approach also yields semantically structured latent spaces, highlighting the value of angular geometry for interpretable and effective representation-based PU learning in visual domains.",
-      "year": 2026,
-      "score": 0.7778,
-      "source": "keyword_search+arxiv_recs",
-      "contaminated": false
-    },
     {
       "paper_id": "28ef77d35cb3712991f8f125fe06c04b45c1289c",
       "arxiv_id": "2603.21840",
@@ -5807,6 +5799,17 @@
       "score": 0.6111,
       "source": "keyword_search+arxiv_recs",
       "contaminated": false
+    },
+    {
+      "paper_id": "web_009adc9922648ddf",
+      "arxiv_id": null,
+      "title": "Foundation Model Makes Clustering A Better",
+      "abstract": "Foundation Model Makes Clustering A Better Initialization For Cold-Start Active Learning Han Yuan1 and Chuan Hong2 1 Centre for Quantitative Medicine, Duke-NUS Medical School 2 Department of Biostatistics and Bioinformatics, Duke University arXiv:2402.02561v2 [cs.LG] 27 Mar 2024 Abstract. Active learning selects the most informative samples from the unlabelled dataset to annotate in the context of a limited anno- tation budget. While numerous methods have been proposed for subse- quent sample selection based on an initialized model, scant attention has been paid to the indispensable phase of active learning: selecting sam- ples for model cold-start initialization. Most of the previous studies resort to random sampling or naive clustering. However, random sampling is prone to fluctuation, and naive clustering suffers from convergence speed, particularly when dealing with high-dimensional data such as imaging data. In this work, we propose to integrate foundation models with clus- tering methods to select samples for cold-start active learning initial- ization. Foundation models refer to those trained on massive datasets by the self-supervised paradigm and capable of generating informative and compacted embeddings for various downstream tasks. Leveraging these embeddings to replace raw features such as pixel values, clustering quickly converges and identifies better initial samples. For a compre- hensive comparison, we included a classic ImageNet-supervised model to acquire emb",
+      "year": null,
+      "score": 0.2638,
+      "source": "manual_web_r3",
+      "source_url": "https://arxiv.org/pdf/2402.02561",
+      "contaminated": false
     }
   ],
   "Sinopolis-Chengdu.pdf": [