Pith. sign in

REVIEW 1 major objections 1 minor 46 references

A verifier trained on LLM diagnostic rationales improves aspect sentiment triplet extraction by up to 3.53 F1 points as a plug-and-play module.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.3

2026-06-28 22:09 UTC pith:GSWCQKZG

load-bearing objection FiVeD trains a multi-objective verifier on LLM-labeled synthetic errors for ASTE, but the lack of any check on those labels is the load-bearing assumption. the 1 major comments →

arxiv 2605.31446 v1 pith:GSWCQKZG submitted 2026-05-29 cs.CL cs.AI

Fine-grained Verification via Diagnostic Reasoning Supervision for Aspect Sentiment Triplet Extraction

Wenna Lai , Haoran Xie , Guandong Xu , Qing Li , S. Joe Qin This is my paper
classification cs.CL cs.AI
keywords aspect sentiment triplet extractionfine-grained verificationdiagnostic reasoning supervisionquality score estimationerror type classificationpost-hoc verificationLLM-generated supervision
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper seeks to establish that post-hoc verification of extracted aspect-opinion-sentiment triplets can be strengthened by training a verifier on fine-grained diagnostic signals rather than simple validity labels. This addresses the problem that current ASTE systems output triplets which appear locally correct yet prove globally invalid, limiting their value for opinion mining, recommendations, and summarization. FiVeD defines hierarchical error categories, builds synthetic incorrect triplets under semantic and syntactic constraints, and uses an off-the-shelf LLM with rubrics to generate quality scores and explanatory rationales. The verifier is then trained on four complementary tasks: validity classification and quality scoring as primary objectives, plus error-type classification and rationale generation as auxiliaries. Experiments show the module raises performance when added to existing extractors and enables tunable precision-recall filtering at inference time.

Core claim

FiVeD is a framework for fine-grained verification with diagnostic reasoning supervision. The verifier is trained with multiple complementary objectives, including validity classification and quality score estimation as primary tasks, with error type classification and rationale generation as auxiliary tasks. Hierarchical error categories are defined and plausible incorrect triplets are constructed under semantic and syntactic constraints. An off-the-shelf LLM with task-specific rubrics produces the quality scores and diagnostic rationales used as supervision. During inference the resulting quality scores filter candidate outputs and support adjustable precision-recall tradeoffs.

What carries the argument

The multi-objective verifier trained on LLM-generated quality scores and diagnostic rationales for validity classification, quality estimation, error typing, and rationale generation.

Load-bearing premise

The off-the-shelf LLM with task-specific rubrics produces reliable quality scores and diagnostic rationales that can serve as supervision for training the verifier.

What would settle it

A direct comparison of LLM-generated quality scores and rationales against human expert annotations on a held-out set of triplets, measuring agreement rates on validity and error types.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • FiVeD raises F1 scores by up to 3.53 points when added to multiple existing ASTE baselines.
  • Quality scores enable filtering that trades precision against recall in a controllable way.
  • The same module works across diverse extractors without retraining them.
  • Hierarchical error categories capture the multi-faceted ways triplets can be invalid.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The diagnostic-supervision approach could transfer to other structured prediction tasks that output tuples needing validation.
  • Systematic biases in the LLM rationales could be inherited by the trained verifier and affect downstream applications.
  • Joint training of extractor and verifier might remove the need for a separate post-hoc stage.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 1 minor

Summary. The paper proposes FiVeD, a framework for fine-grained verification of Aspect Sentiment Triplet Extraction (ASTE) outputs. The verifier is trained on multiple tasks: validity classification and quality score estimation as primary tasks, and error type classification and rationale generation as auxiliary tasks. Synthetic invalid triplets are generated under semantic and syntactic constraints, and an off-the-shelf LLM with task-specific rubrics is used to assign quality scores and diagnostic rationales for supervision. At inference, the quality scores are used to filter candidate triplets from various extractors, allowing adjustable precision-recall tradeoffs. Experiments show that FiVeD improves performance by up to 3.53 F1 points across multiple ASTE baselines as a plug-and-play module.

Significance. If the central claim holds, this work addresses an underexplored gap in post-hoc verification for ASTE, which is relevant for reliability in downstream tasks such as opinion mining and review summarization. The plug-and-play design allows immediate application to existing extractors, and the multi-task diagnostic supervision offers a structured way to handle multi-faceted invalidity and graded usability of triplets.

major comments (1)
  1. [Abstract (paragraph on data construction)] Abstract (paragraph on data construction): The verifier's primary tasks (validity classification, quality score estimation) and auxiliary tasks (error type classification, rationale generation) are all trained on signals from an off-the-shelf LLM with task-specific rubrics. No human validation, inter-annotator agreement, or error analysis of these LLM outputs is referenced. This is load-bearing for the central claim, because if the LLM systematically misclassifies error types or inflates quality scores, the resulting verifier cannot be guaranteed to filter or re-rank on genuine grounds rather than LLM artifacts.
minor comments (1)
  1. [Abstract] The abstract states performance gains of up to 3.53 F1 but supplies no experimental details on baselines, datasets, statistical significance testing, or ablation results. Adding these would strengthen assessment of the plug-and-play claim.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback highlighting a key aspect of our supervision pipeline. We address the major comment below and will revise the manuscript accordingly to strengthen the claims.

read point-by-point responses

  1. Referee: The verifier's primary tasks (validity classification, quality score estimation) and auxiliary tasks (error type classification, rationale generation) are all trained on signals from an off-the-shelf LLM with task-specific rubrics. No human validation, inter-annotator agreement, or error analysis of these LLM outputs is referenced. This is load-bearing for the central claim, because if the LLM systematically misclassifies error types or inflates quality scores, the resulting verifier cannot be guaranteed to filter or re-rank on genuine grounds rather than LLM artifacts.

    Authors: We agree that the absence of human validation for the LLM-generated labels represents a limitation in the current manuscript, as these signals are indeed central to training the verifier. The rubrics were designed to be task-specific and to mitigate common LLM biases (e.g., by requiring explicit justification for quality scores and error types), and the synthetic invalid triplets were generated under explicit semantic/syntactic constraints to reduce hallucination risks. However, without reported human checks, the reliability cannot be fully substantiated. In the revised manuscript, we will add a new subsection under Data Construction describing a human validation study: two independent annotators will evaluate a random sample of 300 LLM-labeled instances (stratified across validity, quality scores, error types, and rationales), reporting Cohen's kappa for inter-annotator agreement and agreement rates with the LLM outputs, along with qualitative error analysis. This will directly address the concern and provide evidence that the supervision is not merely capturing LLM artifacts. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical plug-and-play verifier trained on external LLM signals

full rationale

The paper describes a verification framework (FiVeD) whose primary and auxiliary training objectives are defined over LLM-generated quality scores, rationales, and error labels on synthetically constructed invalid triplets. No equations, derivations, or parameter-fitting steps are present that reduce a claimed prediction back to the input labels by construction. The reported F1 gains are presented as experimental outcomes rather than forced by any self-referential loop. No self-citation chains, uniqueness theorems, or ansatz smuggling appear in the provided text. The method is therefore self-contained against external benchmarks for the purpose of circularity analysis.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework rests on the assumption that hierarchical error categories can be defined and that an LLM can generate usable supervision signals; no free parameters or invented entities are mentioned in the abstract.

axioms (2)
  • domain assumption Hierarchical error categories can be defined for invalid ASTE triplets under semantic and syntactic constraints.
    Used to construct plausible incorrect triplets for training.
  • domain assumption An off-the-shelf LLM supplied with task-specific rubrics can produce reliable quality scores and diagnostic rationales.
    This data is used to train the verifier's primary and auxiliary tasks.

pith-pipeline@v0.9.1-grok · 5784 in / 1327 out tokens · 26254 ms · 2026-06-28T22:09:07.728989+00:00 · methodology

0 comments
read the original abstract

Aspect Sentiment Triplet Extraction (ASTE) aims to identify aspect terms, opinion terms, and sentiment polarities as structured triplets, providing essential inputs for downstream information system applications such as opinion mining, explainable recommendations, and review summarization. Prior work mainly focuses on end-to-end extraction, while post hoc verification of extracted triplets remains comparatively underexplored. This gap limits the reliability of ASTE systems, since predicted triplets may be locally plausible while being globally invalid. Moreover, candidate invalidity is multi-faceted and candidate usability is inherently graded, motivating a fine-grained verification mechanism that can filter or re-rank outputs from diverse extractors. In this paper, we propose FiVeD, a framework for Fine-grained Verification with Diagnostic reasoning supervision. Specifically, the verifier is trained with multiple complementary objectives, including validity classification and quality score estimation as primary tasks, with error type classification and rationale generation as auxiliary tasks. We define hierarchical error categories and construct plausible incorrect triplets under semantic and syntactic constraints, and leverage an off-the-shelf LLM with task-specific rubrics to produce quality scores and diagnostic rationales. During inference, the resulting quality scores are used to filter candidate outputs, supporting adjustable precision-recall tradeoffs. Experiments across multiple ASTE baselines demonstrate that FiVeD consistently improves extraction performance by up to 3.53 F1 points as a plug-and-play verification module.

Figures

Figures reproduced from arXiv: 2605.31446 by Guandong Xu, Haoran Xie, Qing Li, S. Joe Qin, Wenna Lai.

Figure 1
Figure 1. Figure 1: Motivating example of fine-grained triplet verification. An ASTE model may produce both correct triplets and plausible invalid [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of FiVeD for fine-grained ASTE verification. The framework first constructs diagnostic supervision by mining [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: F1 performance under different numbers of verification paths across three generative ASTE pipelines. The dotted line denotes [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Precision-recall curves on the test set for the unified verifier applied to the MVP method with [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Precision-recall curves on the test set for the unified verifier applied to the MVP method with [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Precision-recall curves on the test set for the unified verifier applied to the GAS method with [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Precision-recall curves on the test set for the unified verifier applied to the GAS method with [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Precision-recall curves on the dev and test set for the unified verifier applied to the Paraphrase method with [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Precision-recall curves on the dev and test set for the unified verifier applied to the Paraphrase method with [PITH_FULL_IMAGE:figures/full_fig_p018_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: The automatically learned loss weight assigned to each sub-task with rationale generation from Qwen. [PITH_FULL_IMAGE:figures/full_fig_p019_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: The automatically learned loss weight assigned to each sub-task with rationale generation from DeepSeek-Reasoner. [PITH_FULL_IMAGE:figures/full_fig_p019_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Distribution of constructed quality scores by error type when using Qwen to generate diagnostic reasoning supervision. [PITH_FULL_IMAGE:figures/full_fig_p020_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Distribution of constructed quality scores by error type when using DeepSeek-Reasoner to generate diagnostic reasoning [PITH_FULL_IMAGE:figures/full_fig_p020_13.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

46 extracted references · 10 canonical work pages · 2 internal anchors

  1. [1]

    Aaditya Bodke, Avinoor Singh Kohli, Hemant Subhash Pardeshi, and Prathamesh Bhosale. 2025. PASTEL : Polarity-Aware Sentiment Triplet Extraction with LLM-as-a-Judge. InFindings of the Association for Computational Linguistics: ACL 2025. Association for Computational Linguistics, Vienna, Austria, 25523–25533

    2025

  2. [2]

    Hao Chen, Zepeng Zhai, Fangxiang Feng, Ruifan Li, and Xiaojie Wang. 2022. Enhanced Multi-Channel Graph Convolutional Network for Aspect Sentiment Triplet Extraction. InProceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Dublin, Ireland, 2974–2985

    2022

  3. [3]

    Yujun Chen, Mingwei Tang, Shangyi Du, Kun Yang, Yanxi Zheng, and Mingfeng Zhao. 2026. DGSEP: Dual-stage generative model with sequence- oriented labeling and element-to-tuple prompting improves aspect sentiment triplet extraction.Expert Systems with Applications299 (2026), 130088

    2026

  4. [4]

    Scaling Instruction-Finetuned Language Models

    Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Alex Castro-Ros, Marie Pellat, Kevin Robinson, Dasha Valter, Sharan Narang, Gaurav Mishra, Adams Yu, Vincent Zhao, Yanping H...

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  5. [5]

    Jingfeng Cui, Zhaoxia Wang, Seng-Beng Ho, and Erik Cambria. 2023. Survey on sentiment analysis: evolution of research methods and topics.Artif. Intell. Rev.56, 8 (2023), 8469–8510

    2023

  6. [6]

    Tianhao Gao, Jun Fang, Hanyu Liu, Zhiyuan Liu, Chao Liu, Pengzhang Liu, Yongjun Bao, and Weipeng Yan. 2022. LEGO-ABSA: A Prompt-based Task Assemblable Unified Generative Framework for Multi-task Aspect-based Sentiment Analysis. InProceedings of the 29th International Conference on Computational Linguistics, COLING 2022, Gyeongju, Republic of Korea, Octobe...

    2022

  7. [7]

    Zhibin Gou, Qingyan Guo, and Yujiu Yang. 2023. MvP: Multi-view Prompting Improves Aspect Sentiment Tuple Prediction. InProceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2023, Toronto, Canada, July 9-14, 2023. 4380–4397

    2023

  8. [8]

    Jerry Huang, Siddarth Madala, Risham Sidhu, Cheng Niu, Hao Peng, Julia Hockenmaier, and Tong Zhang. 2026. Tackling Distractor Documents in Multi-Hop QA with Reinforcement and Curriculum Learning. InFindings of the Association for Computational Linguistics: EACL 2026. 5548–5561

    2026

  9. [9]

    Kalervo Järvelin and Jaana Kekäläinen. 2002. Cumulated gain-based evaluation of IR techniques.ACM Transactions on Information Systems20, 4 (Oct. 2002), 422–446

    2002

  10. [10]

    2025.RLSF: Fine-tuning LLMs via Symbolic Feedback

    Piyush Jha, Prithwish Jana, Pranavkrishna Suresh, Arnav Arora, and Vijay Ganesh. 2025.RLSF: Fine-tuning LLMs via Symbolic Feedback. IOS Press. doi:10.3233/faia250996

    work page doi:10.3233/faia250996 2025

  11. [11]

    Wenna Lai, Haoran Xie, Guandong Xu, and Qing Li. 2025. RVISA: Reasoning and Verification for Implicit Sentiment Analysis.IEEE Transactions on Affective Computing16, 3 (2025), 1760–1771. doi:10.1109/TAFFC.2025.3537799 Manuscript submitted to ACM 24 Lai et al

    work page doi:10.1109/taffc.2025.3537799 2025

  12. [12]

    Wenna Lai, Haoran Xie, Guandong Xu, and Qing Li. 2025. STAR: Stepwise Task Augmentation and Relation Learning for Aspect Sentiment Quad Prediction. arXiv:2501.16093 [cs.CL]

    work page arXiv 2025

  13. [13]

    Wenna Lai, Haoran Xie, Guandong Xu, and Qing Li. 2026. Multi-Task Learning With LLMs for Implicit Sentiment Analysis: Data-Level and Task-Level Automatic Weight Learning.IEEE Transactions on Knowledge and Data Engineering38, 01 (Jan. 2026), 506–517

    2026

  14. [14]

    Wenna Lai, Haoran Xie, Guandong Xu, Qing Li, and S. Joe Qin. 2025. Listwise Preference Optimization with Element-wise Confusions for Aspect Sentiment Quad Prediction. arXiv:2511.23184 [cs.CL] https://arxiv.org/abs/2511.23184

    work page arXiv 2025

  15. [15]

    Wenna Lai, Haoran Xie, Guandong Xu, Qing Li, and S. Joe Qin. 2025. When LLMs Team Up: The Emergence of Collaborative Affective Computing. arXiv:2506.01698 [cs.CL] https://arxiv.org/abs/2506.01698

    work page arXiv 2025

  16. [16]

    Dongxu Li, Zhihao Yang, Yuquan Lan, Yunqi Zhang, Hui Zhao, and Gang Zhao. 2023. Simple Approach for Aspect Sentiment Triplet Extraction Using Span-Based Segment Tagging and Dual Extractors(SIGIR ’23). Association for Computing Machinery, New York, NY, USA, 2374–2378

    2023

  17. [17]

    Siheng Li, Cheng Yang, Taiqiang Wu, Chufan Shi, Yuji Zhang, Xinyu Zhu, Zesen Cheng, Deng Cai, Mo Yu, Lemao Liu, Jie Zhou, Yujiu Yang, Ngai Wong, Xixin Wu, and Wai Lam. 2024. A Survey on the Honesty of Large Language Models.Transactions on Machine Learning Research2025 (2024)

    2024

  18. [18]

    You Li, Xupeng Zeng, Yixiao Zeng, and Yuming Lin. 2024. Enhanced Packed Marker with Entity Information for Aspect Sentiment Triplet Extraction. InProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval(Washington DC, USA)(SIGIR ’24). Association for Computing Machinery, New York, NY, USA, 619–629

    2024

  19. [19]

    Lukas Liebel and Marco Körner. 2018. Auxiliary Tasks in Multi-task Learning.arXiv(2018)

    2018

  20. [20]

    Hunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. 2024. Let’s Verify Step by Step. InThe Twelfth International Conference on Learning Representations

    2024

  21. [21]

    Bill Yuchen Lin, Ronan Le Bras, Kyle Richardson, Ashish Sabharwal, Radha Poovendran, Peter Clark, and Yejin Choi. 2025. ZebraLogic: On the Scaling Limits of LLMs for Logical Reasoning. InForty-second International Conference on Machine Learning. https://openreview.net/forum?id=sTAJ9QyA6l

    2025

  22. [22]

    Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, Shashank Gupta, Bodhisattwa Prasad Majumder, Katherine Hermann, Sean Welleck, Amir Yazdanbakhsh, and Peter Clark. 2023. Self-Refine: Iterative Refinement with Self-Feedback. InNeurIPS

    2023

  23. [23]

    Yue Mao, Yi Shen, Chao Yu, and Longjun Cai. 2021. A Joint Training Dual-MRC Framework for Aspect Based Sentiment Analysis. InThirty-Fifth AAAI Conference on Artificial Intelligence, AAAI 2021, Thirty-Third Conference on Innovative Applications of Artificial Intelligence, IAAI 2021, The Eleventh Symposium on Educational Advances in Artificial Intelligence,...

    work page doi:10.1609/aaai.v35i15.17597 2021

  24. [24]

    Iwo Naglik and Mateusz Lango. 2024. ASTE-Transformer: Modelling Dependencies in Aspect-Sentiment Triplet Extraction. InFindings of the Association for Computational Linguistics: EMNLP 2024. Association for Computational Linguistics, Miami, Florida, USA, 2324–2339. doi:10.18653/v1/ 2024.findings-emnlp.129

    work page doi:10.18653/v1/ 2024

  25. [25]

    Haiyun Peng, Lu Xu, Lidong Bing, Fei Huang, Wei Lu, and Luo Si. 2020. Knowing What, How and Why: A Near Complete Solution for Aspect-based Sentiment Analysis.The Thirty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2020, The Thirty-Second Innovative Applications of Artificial Intelligence Conference, IAAI 2020, The Tenth AAAI Symposium on Educat...

    2020

  26. [26]

    Kun Peng, Chaodong Tong, Cong Cao, Hao Peng, Qian Li, Guanlin Wu, Lei Jiang, Yanbing Liu, and Philip S. Yu. 2025. T-T: table transformer for tagging-based aspect sentiment triplet extraction. InProceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence(Montreal, Canada)(IJCAI ’25). Article 914, 9 pages

    2025

  27. [27]

    Maria Pontiki, Dimitris Galanis, Haris Papageorgiou, Ion Androutsopoulos, Suresh Manandhar, Mohammad AL-Smadi, Mahmoud Al-Ayyoub, Yanyan Zhao, Bing Qin, Orphée De Clercq, Véronique Hoste, Marianna Apidianaki, Xavier Tannier, Natalia Loukachevitch, Evgeniy Kotelnikov, Nuria Bel, Salud María Jiménez-Zafra, and Gülşen Eryiğit. 2016. SemEval-2016 Task 5: Aspe...

    2016

  28. [28]

    Maria Pontiki, Dimitris Galanis, Haris Papageorgiou, Suresh Manandhar, and Ion Androutsopoulos. 2015. SemEval-2015 Task 12: Aspect Based Sentiment Analysis. InProceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015). Denver, Colorado, 486–495

    2015

  29. [29]

    Maria Pontiki, Dimitris Galanis, John Pavlopoulos, Harris Papageorgiou, Ion Androutsopoulos, and Suresh Manandhar. 2014. SemEval-2014 Task 4: Aspect Based Sentiment Analysis. InSemEval@COLING

    2014

  30. [30]

    Kim Schouten and Flavius Frasincar. 2016. Survey on Aspect-Level Sentiment Analysis.IEEE Transactions on Knowledge and Data Engineering28, 3 (2016), 813–830

    2016

  31. [31]

    Qiao Sun, Liujia Yang, Minghao Ma, Nanyang Ye, and Qinying Gu. 2024. MiniConGTS: A Near Ultimate Minimalist Contrastive Grid Tagging Scheme for Aspect Sentiment Triplet Extraction. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, Miami, Florida, USA, 2817–2834

    2024

  32. [32]

    Yuxiang Wei, Olivier Duchenne, Jade Copet, Quentin Carbonneaux, LINGMING ZHANG, Daniel Fried, Gabriel Synnaeve, Rishabh Singh, and Sida Wang. 2026. SWE-RL: Advancing LLM Reasoning via Reinforcement Learning on Open Software Evolution. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems

    2026

  33. [33]

    Luo Xianlong, Meng Yang, and Yihao Wang. 2023. Tagging-Assisted Generation Model with Encoder and Decoder Supervision for Aspect Sentiment Triplet Extraction. InProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, Singapore, 2078–2093. Manuscript submitted to ACM Fine-grained Ve...

    2023

  34. [34]

    Tian Xie, Zitian Gao, Qingnan Ren, Haoming Luo, Yuqian Hong, Bryan Dai, Joey Zhou, Kai Qiu, Zhirong Wu, and Chong Luo. 2025. Logic-RL: Unleashing LLM Reasoning with Rule-Based Reinforcement Learning. arXiv:2502.14768 [cs.CL]

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  35. [35]

    Lu Xu, Yew Ken Chia, and Lidong Bing. 2021. Learning Span-Level Interactions for Aspect Sentiment Triplet Extraction. InProceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). Association for Computational Linguistics, Online...

    2021

  36. [36]

    Lu Xu, Hao Li, Wei Lu, and Lidong Bing. 2020. Position-Aware Tagging for Aspect Sentiment Triplet Extraction. InProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP). Association for Computational Linguistics, Online, 2339–2349

    2020

  37. [37]

    Tianyang Xu, Shujin Wu, Shizhe Diao, Xiaoze Liu, Xingyao Wang, Yangyi Chen, and Jing Gao. 2024. SaySelf: Teaching LLMs to Express Confidence with Self-Reflective Rationales. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. Miami, Florida, USA, 5985–5998

    2024

  38. [38]

    Hang Yan, Junqi Dai, Tuo Ji, Xipeng Qiu, and Zheng Zhang. 2021. A Unified Generative Framework for Aspect-based Sentiment Analysis. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). Association for Computational Ling...

    2021

  39. [39]

    Zepeng Zhai, Hao Chen, Fangxiang Feng, Ruifan Li, and Xiaojie Wang. 2022. COM-MRC: A COntext-Masked Machine Reading Comprehension Framework for Aspect Sentiment Triplet Extraction. InProceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, Abu Dhabi, United Arab Emirates, 3230–3241

    2022

  40. [40]

    Chong Zhang, Yue Deng, Xiang Lin, Bin Wang, Dianwen Ng, Hai Ye, Xingxuan Li, Yao Xiao, Zhanfeng Mo, Qi Zhang, and Lidong Bing. 2025. 100 Days After DeepSeek-R1: A Survey on Replication Studies and More Directions for Reasoning Language Models.ArXivabs/2505.00551 (2025). https://api.semanticscholar.org/CorpusID:278237268

    work page Pith review arXiv 2025

  41. [41]

    Wenxuan Zhang, Yang Deng, Xin Li, Yifei Yuan, Lidong Bing, and Wai Lam. 2021. Aspect Sentiment Quad Prediction as Paraphrase Generation. InProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, EMNLP 2021, Virtual Event / Punta Cana, Dominican Republic, 7-11 November, 2021. 9209–9219

    2021

  42. [42]

    Wenxuan Zhang, Yue Deng, Bing Liu, Sinno Pan, and Lidong Bing. 2024. Sentiment Analysis in the Era of Large Language Models: A Reality Check. InFindings of the Association for Computational Linguistics: NAACL 2024. Mexico City, Mexico, 3881–3906

    2024

  43. [43]

    Wenxuan Zhang, Xin Li, Yang Deng, Lidong Bing, and Wai Lam. 2021. Towards Generative Aspect-Based Sentiment Analysis. InProceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, ACL/IJCNLP 2021, (Volume 2: Short Papers), Virtual Event, August 1-6, 202...

    2021

  44. [44]

    Xiaowei Zhao, Yong Zhou, and Xiujuan Xu. 2024. Dual Encoder: Exploiting the Potential of Syntactic and Semantic for Aspect Sentiment Triplet Extraction. InProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). ELRA and ICCL, Torino, Italia, 5401–5413

    2024

  45. [45]

    Xinyu Zhu, Junjie Wang, Lin Zhang, Yuxiang Zhang, Yongfeng Huang, Ruyi Gan, Jiaxing Zhang, and Yujiu Yang. 2023. Solving Math Word Problems via Cooperative Reasoning induced Language Models. InProceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computational Linguistics, Toronto,...

    2023

  46. [46]

    Wang Zou, Wubo Zhang, Wenhuan Wu, and Zhuoyan Tian. 2024. A Multi-task Shared Cascade Learning for Aspect Sentiment Triplet Extraction Using BERT-MRC.Cognitive Computation16 (2024), 1554 – 1571. https://api.semanticscholar.org/CorpusID:267328915 Manuscript submitted to ACM

    2024