title stringlengths 20 147 ⌀ | authors stringlengths 8 1.79k ⌀ | accepted_at stringclasses 1
value | abstract stringlengths 398 2.29k ⌀ | cycle stringclasses 2
values | mean_overall_assessment float64 1 5 ⌀ | overall_assessment listlengths 0 6 ⌀ | mean_confidence float64 1 5 ⌀ | confidence listlengths 0 6 ⌀ | paper_summary listlengths 0 6 ⌀ | summary_of_strengths listlengths 0 6 ⌀ | summary_of_weaknesses listlengths 0 6 ⌀ | comments_suggestions_and_typos listlengths 0 6 ⌀ | soundness listlengths 0 6 ⌀ | excitement listlengths 0 5 ⌀ | datasets listlengths 0 6 ⌀ | software listlengths 0 6 ⌀ | metareview_score float64 2 5 ⌀ | metareview stringlengths 83 6.3k ⌀ | summary_of_reasons_to_publish stringlengths 2 2.54k ⌀ | text_v1 stringlengths 19.7k 627k ⌀ | text_v2 stringlengths 21.3k 635k ⌀ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Learning from Negative Samples in Biomedical Generative Entity Linking | Chanhwi Kim, Hyunjae Kim, Sihyeon Park, Jiwoo Lee, Mujeen Sung, Jaewoo Kang | acl25 | Generative models have become widely used in biomedical entity linking (BioEL) due to their excellent performance and efficient memory usage. However, these models are usually trained only with positive samples—entities that match the input mention’s identifier—and do not explicitly learn from hard negative samples, wh... | aclweb.org/ACL/ARR/2024/December | 2.75 | [
2.5,
3
] | 3 | [
3,
3
] | [
"This paper proposes the incorporation of negative samples in biomedical generative entity linking. The approach is based on two stages:\n\n1. Training with positive samples by learning to generate synonyms using existing knowledge bases as reference.\n2. Training with negative samples by using the DPO algorithm an... | [
"The paper includes a comparison with a number of baselines and state of the art systems, and the results of the proposed system are marginally better.",
"A novel framework to integrate negative samples into generative BioEL, addressing a critical limitation of existing approaches.\nApplicable to both pre-trainin... | [
"1. Novelty is unclear. DPO is used in related work, e.g. Rafailov et al. cited in the paper, to help aligning LLMs to the chosen task.\n\n2. The approach to select most similar synonyms is unclear and questionable. Does \"tf.idf similarity\" mean cosine similarity of the tf.idf of each synonym? If so, why use tf.i... | [
"1. Update the section on related work to refer to DPO, what it was designed for, and how it is used nowadays.\n\n2. Be more explicit on what is the novelty of your approach.",
"Suggestions:\n1. Clarify the impact of UMLS version differences in MedMentions results.\n2. Include top-5 accuracy metrics to assess bro... | [
2.5,
3.5
] | [] | [
1,
1
] | [
2,
1
] | 4 | This paper introduces a method (ANGEL) for biomedical entity-linking that uses negative samples. | The performance improvement is 1.7% across 5 datasets. The method is simple yet effective. Experiments are detailed, with a large number of ablation studies. | Learning from Negative Samples in Biomedical Generative Entity Linking
Anonymous ACL submission
Abstract
Generative models have become widely used in biomedical entity linking (BioEL) due to their excellent performance and efficient memory usage. However, these models are usually trained only with positive samples—e... | Learning from Negative Samples in Biomedical Generative Entity Linking
Chanhwi Kim1*, Hyunjae Kim1*, Sihyeon Park1, Jiwoo Lee1,
Mujeen Sung2†, Jaewoo Kang1,3‡
1Korea University, 2Kyung Hee University, 3AIGEN Sciences
{chanhwi_kim, hyunjae-kim, sh10, hijiwoo7}@korea.ac.kr
mujeensung@khu.ac.kr, kangj@korea.ac.kr
Abstr... |
CRUXEVAL-X: A Benchmark for Multilingual Code Reasoning, Understanding and Execution | Ruiyang Xu, Jialun CAO, Yaojie Lu, Hongyu Lin, Xianpei Han, Ben He, Shing-Chi Cheung, Le Sun | acl25 | "Code benchmarks such as HumanEval are widely adopted to evaluate Large Language Models' (LLMs) codi(...TRUNCATED) | aclweb.org/ACL/ARR/2024/December | 2.75 | [
3,
2.5
] | 4.5 | [
4,
5
] | ["This paper propose a multilingual code reasoning corpus CruxEval-X. The authors provided an automa(...TRUNCATED) | ["1. The proposed corpus is useful in studying code abilities of different LLMs.\n\n2. The automated(...TRUNCATED) | ["1. The dataset is constructed from translating existing Python data to other programming languages(...TRUNCATED) | ["1. The input/output length mentioned in Section 4.1 sounds a little bit weird to me. What is the l(...TRUNCATED) | [
3.5,
3.5
] | [] | [
4,
4
] | [
1,
1
] | 3 | "1. This paper has proposed a multilingual benchmark that expands on existing Python-dominated datas(...TRUNCATED) | See 1&2 in Meta review. | "CRUXEVAL-X: A Benchmark for Multilingual Code Reasoning, Understanding and Execution\n\nAnonymous A(...TRUNCATED) | "CRUXEVAL-X: A Benchmark for Multilingual Code Reasoning, Understanding and Execution\n\nRuiyang Xu1(...TRUNCATED) |
Mind Your Theory: Theory of Mind Goes Deeper Than Reasoning | Eitan Wagner, Nitay Alon, Joseph M Barnby, Omri Abend | acl25 | "Theory of Mind (ToM) capabilities in LLMs have recently become a central object of investigation, s(...TRUNCATED) | aclweb.org/ACL/ARR/2025/February | 2.75 | [
2.5,
3
] | 3.5 | [
4,
3
] | ["The authors cite two-step theories of theory of mind (ToM) capabilities in humans, according to wh(...TRUNCATED) | ["- The paper is interesting and its arguments are reasonably laid out. As it is a short paper, it d(...TRUNCATED) | ["- I'm not entirely convinced of the strict need to separate the DoM from the inference step, at le(...TRUNCATED) | ["- The authors might be more interested in the CogSci conference, or a workshop on computational co(...TRUNCATED) | [
2,
3.5
] | [
2.5,
3
] | [
1,
1
] | [
1,
1
] | 2.5 | "This paper is a position paper about studying theory of mind in LMs, drawing on ideas from cognitiv(...TRUNCATED) | "The reviewers and I agree that there is a core interesting idea here about evaluating ToM and separ(...TRUNCATED) | "Mind Your Theory: Theory of Mind Goes Deeper Than Reasoning\n\nAnonymous ACL submission\n\nAbstract(...TRUNCATED) | "Mind Your Theory: Theory of Mind Goes Deeper Than Reasoning\n\nEitan Wagner*1 Nitay Alon*1,2 (...TRUNCATED) |
Brevity is the soul of sustainability: Characterizing LLM response lengths | Paramita Koley, Soham Poddar, Janardan Misra, Niloy Ganguly, Saptarshi Ghosh | acl25 | "Developing energy-efficient methods for inference is crucial, as a significant portion of the energ(...TRUNCATED) | aclweb.org/ACL/ARR/2024/December | 2.5 | [
2.5
] | 4 | [
4
] | ["This paper studies the generation lengths of state-of-the-art LLMs on short-answer-preferred probl(...TRUNCATED) | ["- The output length is a critical problem in efficient LLM serving, which, if well-aligned with hu(...TRUNCATED) | ["- Lack of in-depth analysis and contribution: To me, most of the findings that this paper represen(...TRUNCATED) | ["I suggest the authors to discuss more in-depth intuitions and empirical analysis on the discrepanc(...TRUNCATED) | [
3
] | [] | [
1
] | [
1
] | 4 | "This paper examines the relationship between output length and energy efficiency in large language (...TRUNCATED) | "- The paper addresses the important issue of optimizing LLM output efficiency to reduce energy cons(...TRUNCATED) | "Brevity is the soul of sustainability: Characterizing LLM response lengths\n\nAnonymous ACL submiss(...TRUNCATED) | "Brevity is the soul of sustainability: Characterizing LLM response lengths\n\nSoham Poddar1, Parami(...TRUNCATED) |
FINECITE: A Novel Approach on Fine-Grained Citation Context Analysis | Lasse M. Jantsch, Dong Jae Koh, Seonghwan Yoon, Jisu Lee, Anne Lauscher, Young-Kyoon Suh | acl25 | "Citation context analysis (CCA) is a field of research studying the role and purpose of citation in(...TRUNCATED) | aclweb.org/ACL/ARR/2025/February | 3.166667 | [
2,
4,
3.5
] | 4.333333 | [
5,
4,
4
] | ["The paper proposes a finetuning methodology for more accurate downstream standard citation classif(...TRUNCATED) | ["The general idea of finetuning models on annotated datasets on finer context segment-based classes(...TRUNCATED) | ["I would like to point out to some technical issues that I think are very important to be addressed(...TRUNCATED) | ["Abstract: It would be good to introduce what is meant by \"elemental dimensions\".","The compariso(...TRUNCATED) | [
2.5,
3,
4
] | [
4,
4,
3.5
] | [
5,
4,
4
] | [
5,
4,
4
] | 2.5 | "This paper proposes FINECITE, a fine-grained citation context extraction and classification framewo(...TRUNCATED) | "The paper proposed a novel technique (via an annotated dataset) of considering context for better (...TRUNCATED) | "FINECITE: A Novel Approach on Fine-Grained Citation Context Analysis\n\nAnonymous ACL submission\n\(...TRUNCATED) | "FINECITE: A Novel Approach For Fine-Grained Citation Context Analysis\n\nLasse Jantsch1 Dong-Jae Ko(...TRUNCATED) |
MLDebugging: Towards Benchmarking Code Debugging Across Multi-Library Scenarios | "JinYang Huang, Xiachong Feng, Qiguang Chen, Hanjie Zhao, Zihui Cheng, Jiesong Bai, Jingxuan Zhou, M(...TRUNCATED) | acl25 | "Code debugging is a crucial task in software engineering, which attracts increasing attention. Whil(...TRUNCATED) | aclweb.org/ACL/ARR/2025/February | 3.125 | [
3,
2.5,
3.5,
3.5
] | 3.75 | [
4,
4,
3,
4
] | ["This paper introduces a novel benchmark MLDebugging designed to evaluate code debugging capabiliti(...TRUNCATED) | ["1. The benchmark addresses the previously unexplored issue of debugging code involving multiple li(...TRUNCATED) | ["1. The paper lacks sufficient detail regarding the selection criteria of the 126 libraries. Inform(...TRUNCATED) | ["See the weaknesses part.","Here are some minor errors I found: \n1. The parameter classification i(...TRUNCATED) | [
4,
2.5,
3.5,
3.5
] | [
4,
2.5,
2,
3
] | [
4,
3,
4,
4
] | [
4,
1,
1,
1
] | 3 | "This paper introduces MLDebugging, a novel benchmark for evaluating LLMs' ability to debug multi-li(...TRUNCATED) | "+ Addresses the critical and underexplored challenge of multi-library debugging.\n+ Proposes a care(...TRUNCATED) | "MLDebugging: Towards Benchmarking Code Debugging Across Multi-Library Scenarios\n\nAnonymous ACL su(...TRUNCATED) | "MLDebugging: Towards Benchmarking Code Debugging Across Multi-Library Scenarios\n\nJinyang Huang♣(...TRUNCATED) |
QG-SMS: Enhancing Test Item Analysis via Student Modeling and Simulation | Bang Nguyen, Tingting Du, Mengxia Yu, Lawrence Angrave, Meng Jiang | acl25 | "While the Question Generation (QG) task has been increasingly adopted in educational assessments, i(...TRUNCATED) | aclweb.org/ACL/ARR/2025/February | 3.5 | [
3.5
] | 5 | [
5
] | ["The authors developed a LLM-based pipeline to evaluate four aspects of multiple-choice test items (...TRUNCATED) | ["- The paper is well-written and easy to follow.\n- The paper embraces the multi-dimensionality of (...TRUNCATED) | ["### Necessity of LLM-based evaluation step\n\nSection 3 states that in step 3, LLMs are used to as(...TRUNCATED) | ["### Typesetting issues and typos\n\n- The closing curly bracket on line 177 seems like it needs to(...TRUNCATED) | [
3
] | [
4
] | [
1
] | [
1
] | null | null | null | "QG-SMS: Enhancing Test Item Analysis via Student Modeling and Simulation\n\nAnonymous ACL submissio(...TRUNCATED) | "QG-SMS: Enhancing Test Item Analysis via Student Modeling and Simulation\n\nBang Nguyen1 Tingtin(...TRUNCATED) |
Improve Vision Language Model Chain-of-thought Reasoning | "Ruohong Zhang, Bowen Zhang, Yanghao Li, Haotian Zhang, Zhiqing Sun, Zhe Gan, Yinfei Yang, Ruoming P(...TRUNCATED) | acl25 | "Chain-of-thought (CoT) reasoning in vision language models (VLMs) is crucial for improving interpre(...TRUNCATED) | aclweb.org/ACL/ARR/2024/December | 3 | [
3,
3
] | 3.5 | [
3,
4
] | ["This work enhances the reasoning capabilities of VLLM through the following pipeline: (1) leveragi(...TRUNCATED) | ["- The proposed reasoning datasets including 193k examples may be useful for the community.\n- The (...TRUNCATED) | ["- The primary concern of this paper is the concept of utilizing distilled rationales and technique(...TRUNCATED) | [
"See above.",
"See weakness 2"
] | [
4,
3
] | [] | [
4,
4
] | [
4,
3
] | 4 | "This work introduces a COT VL dataset constructed by a GPT-based automatic method. The authors the(...TRUNCATED) | "1.This paper constructs a 193k CoT VL dataset, which is beneficial for the community. 2.Extensive e(...TRUNCATED) | "Improve Vision Language Model Chain-of-thought Reasoning\n\nAnonymous ACL submission\n\nAbstract\nC(...TRUNCATED) | "Improve Vision Language Model Chain-of-thought Reasoning\n\nRuohong Zhang*, Bowen Zhang†, Yanghao(...TRUNCATED) |
TAGExplainer: Narrating Graph Explanations for Text-Attributed Graph Learning Models | Bo Pan, Zhen Xiong, Guanchen Wu, Zheng Zhang, Yifei Zhang, Yuntong Hu, Liang Zhao | acl25 | "Representation learning of Text-Attributed Graphs (TAGs) has garnered significant attention due to (...TRUNCATED) | aclweb.org/ACL/ARR/2024/December | 4 | [
4
] | 2 | [
2
] | ["This paper introduces TAGExplainer, a novel method designed to generate natural language explanati(...TRUNCATED) | ["1. The topic and idea are interesting.\n\n1. The proposed TAGExplainer framework is the first to a(...TRUNCATED) | ["I am not an expert in the text-attributed graph learning field and I do not identify important wea(...TRUNCATED) | ["The authors provided a comprehensive review of natural language explanation in the related works a(...TRUNCATED) | [
4
] | [] | [
3
] | [
3
] | 4 | "AGExplainer is a novel, model-agnostic framework developed to generate natural language explanation(...TRUNCATED) | "- (3D68, Cn3F) The paper addresses a critical yet underexplored area of explainability in Text-Attr(...TRUNCATED) | "TAGExplainer: Narrating Graph Explanations for Text-Attributed Graph Learning Models\n\nAnonymous A(...TRUNCATED) | "GraphNarrator: Generating Textual Explanations for Graph Neural Networks\n\nBo Pan*, Zhen Xiong*, G(...TRUNCATED) |
"CoreEval: Automatically Building Contamination-Resilient Datasets with Real-World Knowledge toward (...TRUNCATED) | Jingqian Zhao, Bingbing Wang, Geng Tu, Yice Zhang, Qianlong Wang, Bin Liang, Jing Li, Ruifeng Xu | acl25 | "Data contamination poses a significant challenge to the fairness of LLM evaluations in natural lang(...TRUNCATED) | aclweb.org/ACL/ARR/2025/February | 3.5 | [
3.5
] | 4 | [
4
] | ["To improve the current research for mitigating data contamination, this work proposes a method ins(...TRUNCATED) | ["The proposed method is a more informed way of data augmentation instead of relying on a single pro(...TRUNCATED) | ["- The presentation of the paper makes it difficult to understand some of the important parts of th(...TRUNCATED) | ["1. Sections 3.2 and 3.3 can be simplified when explaining the steps of the algorithm. For example,(...TRUNCATED) | [
3.5
] | [
3
] | [
1
] | [
1
] | 3.5 | "This paper proposes an approach called CoreEval to mitigate the problem of data contamination durin(...TRUNCATED) | "1.\tThe proposed CoreEval framework introduces a three-stage process that integrates real-world kno(...TRUNCATED) | "CoreEval: Automatically Building Contamination-Resilient Datasets with Real-World Knowledge toward (...TRUNCATED) | "CoreEval: Automatically Building Contamination-Resilient Datasets with Real-World Knowledge toward (...TRUNCATED) |
End of preview. Expand in Data Studio
ACL ARR Reviews - Review Arcade Project
This dataset contains paper reviews from the ACL ARR (Association for Computational Linguistics - Annual Review of Research) program. The reviews are organized into two splits: accepted and rejected, corresponding to papers that were accepted or rejected for publication, respectively.
Record Format
Each line in the JSONL files represents one paper review record. The fields below describe the expected schema.
| Field | Type | Description |
|---|---|---|
title |
string | Paper title. |
authors |
string | Author list as a single string. |
accepted_at |
string | Venue or event where the paper was accepted (for example acl25). |
abstract |
string | Paper abstract text. |
cycle |
string | Review cycle identifier/source path (for example aclweb.org/ACL/ARR/2024/December). |
mean_overall_assessment |
number | Mean of reviewer overall-assessment scores. |
overall_assessment |
array[number] | Raw reviewer overall-assessment scores. |
mean_confidence |
number | Mean of reviewer confidence scores. |
confidence |
array[number] | Raw reviewer confidence scores. |
paper_summary |
array[string] | Reviewer-written summaries of the paper. |
summary_of_strengths |
array[string] | Reviewer-listed strengths. |
summary_of_weaknesses |
array[string] | Reviewer-listed weaknesses. |
comments_suggestions_and_typos |
array[string] | Reviewer comments, suggestions, and typo/edit notes. |
soundness |
array[number] | Reviewer soundness scores. |
excitement |
array[number] | Reviewer excitement scores (can be empty). |
datasets |
array[number] | Reviewer dataset-related scores. |
software |
array[number] | Reviewer software/reproducibility scores. |
metareview_score |
number | Meta-review score. |
metareview |
string | Meta-review summary text. |
summary_of_reasons_to_publish |
string | Meta-review rationale for publication. |
text_v1 |
string | Full paper/review text snapshot (version 1). |
text_v2 |
string | Full paper/review text snapshot (version 2, camera-ready version). |
- Downloads last month
- 9