Pith. sign in

REVIEW 3 major objections 5 minor 85 references

How you ask an LLM to find emotion–cause pairs matters more than the model: judging one pair at a time beats generating the full set.

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 ↗

Pair-level judgement consistently outperforms dialogue-level generation for LLM-based ECPEC because models recognize pairs under explicit queries but fail at set-level discovery and shared-threshold decisions.

T0 review reviewed 2026-07-30 challenge →

load-bearing objection Solid controlled finding that pair judgement beats full-dialogue generation for LLM ECPEC; the set-return bottleneck story is useful, but unequal LoRA setups weaken pure-formulation attribution. the 3 major comments →

arxiv 2607.26967 v1 pith:3WD7RWXI submitted 2026-07-29 cs.CL

Generation or Judgement? A Paradigm Perspective on LLM-Based Emotion-Cause Pair Extraction in Conversation

Weijie Feng , Hongchuang Wang , Binbin Liu , Zhiyong Cheng This is my paper
classification cs.CL
keywords emotion-cause pair extractionconversationlarge language modelstask formulationpair-level judgementdialogue-level generationcandidate scopedecision boundary
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 reading

This paper asks a practical question about using large language models for emotion–cause pair extraction in conversation: does it matter whether the model is asked to dump every pair from a whole dialogue, pick causes for one target utterance, or answer yes/no on a single candidate pair? Across controlled comparisons, pair-level judgement consistently beats dialogue-level generation. The authors show the gap is not mainly that models cannot see the relations—when queried pair by pair they recognize the vast majority of gold links—but that generating a complete set at once is hard, especially when dialogues have many pairs, targets have multiple causes, or causes sit farther away. Judgement eases that discovery burden, yet a single shared yes/no threshold still fails to turn good rankings into clean sets. A light auxiliary retriever that rechecks only ambiguous boundary cases improves F1 a little on three datasets without blowing up cost, arguing that task decomposition and candidate scope are the levers that actually unlock LLMs on this task.

Core claim

Task formulation substantially shapes LLM performance on emotion–cause pair extraction in conversation: pair-level judgement outperforms dialogue-level generation in all 18 controlled backbone–input–scope comparisons, because models can recognize most gold relations under explicit pair queries (about 92.7%–98.1% recall) yet struggle to discover and return complete pair sets in one generative pass; rankings under judgement are more reliable than binary decisions from a shared threshold.

What carries the argument

Three controlled inference paradigms by granularity—dialogue-level generation (GEN), target-level selection (SEL), and pair-level judgement (JUD)—plus local versus full candidate scope, used to isolate how task decomposition and candidate range expose or hide pair recognition; a diagnosis-guided auxiliary retriever then re-examines only the highest-ranked rejected boundary candidate.

Load-bearing premise

The claim that the performance gap is purely about how the task is posed rests on treating separately fine-tuned adapters for each paradigm as fair apples-to-apples comparisons, even though their training capacity and schedules differ.

What would settle it

Retrain GEN, SEL, and JUD with matched LoRA rank, alpha, dropout, epochs, and effective capacity on the same backbone and data; if the JUD-over-GEN F1 gap disappears or shrinks sharply across the same 18 settings, the paradigm diagnosis is confounded by optimization rather than task formulation.

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

If this is right

  • For LLM-based ECPEC, defaulting to pair-level judgement over local candidates (self plus up to four preceding utterances) is a stronger practical baseline than one-shot full-dialogue generation.
  • Gains from finer decomposition come mainly from removing set-level discovery burden, not from giving the model new causal knowledge.
  • Shared binary thresholds leave separable rankings unrealized; adapting or correcting decision boundaries is a direct path to better F1.
  • Restricting candidate scope keeps most gold pairs while cutting the O(N²) cost of full judgement toward roughly linear time.
  • A small boundary retriever can trade a modest compute increase (~1.49× baseline time) for consistent F1 lifts without rewriting the main paradigm.

Where Pith is reading between the lines

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

  • The same generation-versus-judgement split may show up in other multi-instance extraction tasks (e.g., event argument sets, multi-hop relation bundles) where models recognize items better than they exhaustively list them.
  • If matched-capacity training still leaves a large gap, prompting and decoding strategies that force complete set coverage become higher priority than bigger backbones alone.
  • Target-specific thresholds or calibrated scores could capture much of the Ideal-vs-Realized separability gap the paper measures without full pairwise re-inference.
  • Multimodal evidence did not close the paradigm gap here, suggesting formulation may dominate modality fusion for this family of conversational pair tasks.
Share X Bluesky LinkedIn Reddit HN

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

3 major / 5 minor

Summary. The paper argues that how ECPEC is posed to LLMs—dialogue-level generation (GEN), target-level selection (SEL), or pair-level judgement (JUD)—substantially changes performance. Under controlled fine-tuning, JUD beats GEN in all 18 backbone–input–scope comparisons on ECF, ConvECPE, and RECCON. Progressive probes show that under explicit pair queries the same backbone recovers 92.7%–98.1% of gold pairs, so GEN’s misses are attributed mainly to set-level discovery and return rather than inability to recognize relations. JUD’s rankings are more reliable than a shared Yes/No threshold (ISR ≫ RSR); a lightweight auxiliary retriever that re-checks the top rejected boundary candidate then yields +0.50–1.46 F1 at 1.49× GEN-Full inference cost. The authors conclude that task decomposition and candidate scope are critical for LLM-based ECPEC.

Significance. If the paradigm gap is truly driven by formulation rather than training confounds, the paper gives a clear, actionable account of how to use LLMs for ECPEC: prefer pair-level queries, keep a local candidate window, and treat shared thresholds as the remaining bottleneck. The multi-dataset, multi-backbone, multimodal, and local/full-scope design, plus the ROR/ISR/RSR diagnostics and efficiency table, are concrete contributions that go beyond another leaderboard entry. The progressive decomposition in §4.2 and the ranking-vs-threshold taxonomy in §5.2 are especially useful for the broader community working on structured extraction with LLMs. Credit is due for reporting inference cost carefully (Table 8) and for a diagnosis-guided remedy rather than an ad-hoc system.

major comments (3)
  1. [Appendix B.4 / Table 11; §3] Appendix B.4 / Table 11 assigns unequal LoRA capacity and schedules by paradigm (JUD often r=16, α=32, dropout 0.05, 2 epochs vs GEN/SEL r=8, α=16, dropout 0, 6 epochs). The central claim that task formulation alone drives the 18-way JUD>GEN gap (§3, abstract) therefore confounds granularity with effective adapter capacity and optimization trajectory. Matched-budget re-training (same r/α/dropout/epochs, or equal trainable parameters and wall-clock) on at least one backbone and one dataset is needed before the pure-formulation attribution is secure. Without it, both the gap and the §4 “set-level bottleneck” diagnosis remain partly training artifacts.
  2. [§4.2; Figure 4] §4.2’s progressive ladder is the strongest internal evidence that set-level prediction is the dominant omission source, but dialogue-level GEN is only an “additional probe” on a checkpoint jointly fine-tuned on the three gold-target formats. Early steps of the ladder therefore mix true set-prediction difficulty with format mismatch relative to the training mixture. Please either (i) report the same ladder on a GEN-only fine-tuned checkpoint, or (ii) clearly qualify that the 30.83–55.04 point jump isolates pair vs set prediction only under the joint gold-target training regime, not under the GEN training used in Table 1.
  3. [Table 1; Table 7; Appendix B.4] All main results are single-run with seed 13 (Appendix B.4). On ConvECPE the paradigm spread is 18–28 F1 points, so the qualitative ordering is unlikely to reverse, but the remedy gains of +0.50–1.46 F1 (Table 7) and several close GEN/SEL/JUD comparisons in Table 1 sit inside typical run-to-run noise for LoRA fine-tunes. At minimum, report 3 seeds for JUD-Local vs GEN-Local on one dataset and for the retriever delta; otherwise the small remedy claim is under-supported relative to the paper’s emphasis on it.
minor comments (5)
  1. [Table 1; §5.2] Table 1: JUD-Full on ConvECPE has very high recall (79.48) and low precision (33.27), yielding F1 below SEL-Full; a one-sentence discussion of when full-scope JUD hurts would help readers choose configurations.
  2. [Figure 1] Figure 1(b) is referenced as showing unrealized separation potential but is hard to read without axis labels or a short caption definition of the shared boundary; align the caption with the ISR/RSR definitions in §5.2.
  3. [§3.1; Table 10] Local window size 4 is justified by Table 10 (87–94% of gold pairs) but is never ablated (e.g., window 2 vs 4 vs 6). A short ablation would strengthen the “controlled candidate scope” recommendation.
  4. [Appendix A] Prompts in Appendix A use [EMOTION LABEL SET] and [SCOPE INSTRUCTION] placeholders; stating the exact label inventories per dataset would aid reproducibility.
  5. [§2; Appendix B.1] Typos/style: “JUDtheoretically” (missing space, §2); “we choice the RECCON-DD” → “choose” (B.1); occasional missing spaces before citations.

Circularity Check

0 steps flagged

No significant circularity: empirical paradigm comparison scored against external gold pair annotations.

full rationale

The paper’s load-bearing claims are empirical method comparisons (GEN vs SEL vs JUD; local vs full scope; an auxiliary boundary retriever), not first-principles derivations. Performance is measured by precision/recall/F1 of predicted emotion–cause utterance-index pairs against held-out gold annotations on public benchmarks (ECF, ConvECPE, RECCON). Recoverable omission rate, ISR/RSR, and progressive-decomposition recall are defined from predicted vs gold sets and from score orderings under explicit pair queries; none equals a fitted objective by construction. Training adapters on the same task labels used for evaluation is ordinary supervised NLP, not circular reduction. Author-overlapping citations (e.g., SCALE / Ma et al. 2026) appear only as competitive baselines, not as uniqueness theorems or load-bearing premises that force the JUD>GEN gap. Unequal LoRA ranks/schedules across paradigms are a possible experimental confound for attributing the gap purely to formulation, but that is a correctness/identification concern, not circularity under the stated criteria. No self-definitional step, fitted-input-as-prediction, or ansatz-via-self-citation chain is present.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 4 invented entities

Load-bearing content is experimental protocol, not formal axioms. The claim rests on standard supervised fine-tuning assumptions, public ECPEC label definitions, a locality prior for candidate scope, a fixed decision threshold, and paradigm-specific LoRA training choices. No new physical entities; invented constructs are operational (three paradigms, ROR/ISR/RSR, boundary retriever).

free parameters (4)
  • JUD Yes-token probability threshold = 0.5
    Binary accept if normalized P(Yes)>0.5; fixed across datasets, not tuned on val/test (Appendix B.4). Directly shapes precision/recall and the shared-boundary failure the paper diagnoses.
  • Local candidate window size = 4 preceding utterances (0≤i−j≤4)
    Causes restricted to target plus up to four preceding utterances, justified by 87.0–93.9% gold coverage (Table 10). Changes recall ceiling and cost; out-of-window gold pairs count as FN under local eval.
  • Paradigm-specific LoRA rank/alpha/epochs/dropout = See Table 11 (e.g., LLaMA JUD r=16 α=32 ep=2; GEN r=8 α=16 ep=6)
    Different adapter capacity and schedule per paradigm (e.g., JUD often 2 epochs, higher rank; GEN/SEL 6 epochs). Chosen for training, not derived; can confound pure paradigm comparisons (Table 11).
  • Boundary retriever: re-examine only highest-ranked rejected candidate = top-1 rejected candidate per target
    Design choice limiting correction to one Case-3-style candidate per target; caps recall recovery (11.71–19.12% of reconsidered golds) and thus F1 gain size (§6).
axioms (5)
  • domain assumption Gold ECPEC annotations on ECF, ConvECPE, and RECCON-DD are a valid external standard for pair correctness.
    All P/R/F1, ROR, and recognition rates are defined against these labels (§3.1, Appendix B.1).
  • domain assumption Emotion–cause causality is adequately modeled as ordered utterance-index pairs, including self-causes and multi-cause targets.
    Task definition Eq. (1) and evaluation as exact index-pair match (Appendix B.4).
  • domain assumption LoRA fine-tuning plus deterministic greedy/single-pass decoding (temperature 0) is a fair way to compare inference paradigms.
    Stated training/inference protocol in Appendix B.4; no decoding-search ablations.
  • ad hoc to paper Distilling prior LLM-ECPEC systems into GEN, SEL, and JUD captures the essential granularity axis of task formulation.
    §2 taxonomy; multi-agent debate and speaker-centric agents are acknowledged then set aside as out of controlled scope (Appendix B.3).
  • standard math Standard arithmetic definitions of precision, recall, F1 over pair sets.
    Equation (5) in Appendix B.4.
invented entities (4)
  • Three controlled paradigms GEN / SEL / JUD no independent evidence
    purpose: Operationalize task-decomposition granularity for matched comparison.
    Defined in §2 and Figure 2; implementations are paper-specific even if inspired by prior systems.
  • Recoverable Omission Rate (ROR) no independent evidence
    purpose: Measure how many GEN false negatives another paradigm recovers.
    Appendix C definitions ROG→S and ROG→J; metric invented for this analysis.
  • Ideal/Realized Separability Ratios (ISR / RSR) and Case-1–4 ranking taxonomy no independent evidence
    purpose: Separate ranking failures from shared-threshold failures under JUD.
    §5.2 / Table 5; internal diagnostic constructs.
  • Auxiliary boundary retriever no independent evidence
    purpose: Selectively re-accept the top rejected candidate to approximate target-specific thresholds.
    §6; fine-tuned from JUD checkpoint on mixture including boundary instances; gains are only shown inside this paper’s pipeline.

reviewed 2026-07-30 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Generation or Judgement? A Paradigm Perspective on LLM-Based Emotion-Cause Pair Extraction in Conversation." pith.science (2026). https://pith.science/paper/3WD7RWXI

@misc{pith2026260726967,
  author       = {Pith},
  title        = {Pith review of: Generation or Judgement? A Paradigm Perspective on LLM-Based Emotion-Cause Pair Extraction in Conversation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3WD7RWXI}},
  note         = {Machine review of arXiv:2607.26967}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Emotion-cause pair extraction in conversation (ECPEC) identifies utterance pairs in which one utterance causes an emotion expressed in another. Recent LLM-based approaches formulate ECPEC at markedly different granularities, ranging from generating complete pair sets to judging individual candidate pairs. In this paper, we make the surprising observation that task formulation substantially affects performance, where pair-level judgement outperforms dialogue-level generation in all 18 controlled comparisons. We investigate the sources of this paradigm gap and find that many relations omitted by dialogue-level generation remain recognizable under explicit pair queries, under which the model recognizes 92.7%-98.1% of emotion-cause relations. This suggests that LLMs can recognize emotion-cause relations but struggle to discover and return complete pair sets. Pair-level judgement alleviates this burden, although its candidate rankings are more reliable than the binary decisions produced by a shared threshold. Based on this diagnosis, we introduce an auxiliary retriever that selectively re-examines ambiguous boundary cases, yielding consistent F1 improvements of 0.50-1.46 points across three datasets while maintaining an inference time of only 1.49x that of the baseline paradigm. These findings show that task decomposition and candidate scope are critical to effectively utilizing LLMs for ECPEC.

Figures

Figures reproduced from arXiv: 2607.26967 by Binbin Liu, Hongchuang Wang, Weijie Feng, Zhiyong Cheng.

Figure 1
Figure 1. Figure 1: Paradigm gap and unrealized separation potential. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the three LLM-based ECPEC paradigms. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Emotion–cause pairs omitted by GEN [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Results under progressive task decomposition. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Pair recall (%) versus emotion–cause distances. [PITH_FULL_IMAGE:figures/full_fig_p005_5.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

85 extracted references · 5 canonical work pages

  1. [1]

    2010 , month = may, publisher =

    Twitter as a Corpus for Sentiment Analysis and Opinion Mining , booktitle =. 2010 , month = may, publisher =

    2010

  2. [2]

    Emotion Cause Detection with Linguistic Constructions , booktitle =

    Chen, Ying and Lee, Sophia Yat Mei and Li, Shoushan and Huang, Chu-Ren , editor =. Emotion Cause Detection with Linguistic Constructions , booktitle =. 2010 , month = aug, pages =

    2010

  3. [3]

    End-to-End Emotion-Cause Pair Extraction with Graph Convolutional Network , booktitle =

    Chen, Ying and Hou, Wenjun and Li, Shoushan and Wu, Caicong and Zhang, Xiaoqiang , editor =. End-to-End Emotion-Cause Pair Extraction with Graph Convolutional Network , booktitle =. 2020 , month = dec, pages =. doi:10.18653/v1/2020.coling-main.17 , url =

    doi:10.18653/v1/2020.coling-main.17 2020

  4. [4]

    Conditional Causal Relationships between Emotions and Causes in Texts , booktitle =

    Chen, Xinhong and Li, Qing and Wang, Jianping , editor =. Conditional Causal Relationships between Emotions and Causes in Texts , booktitle =. 2020 , month = nov, pages =. doi:10.18653/v1/2020.emnlp-main.252 , url =

    doi:10.18653/v1/2020.emnlp-main.252 2020

  5. [5]

    Joint Alignment of Multi-Task Feature and Label Spaces for Emotion Cause Pair Extraction , booktitle =

    Chen, Shunjie and Shi, Xiaochuan and Li, Jingye and Wu, Shengqiong and Fei, Hao and Li, Fei and Ji, Donghong , editor =. Joint Alignment of Multi-Task Feature and Label Spaces for Emotion Cause Pair Extraction , booktitle =. 2022 , month = oct, pages =

    2022

  6. [6]

    End-to-End Emotion-Cause Pair Extraction Based on Sliding Window Multi-Label Learning , booktitle =

    Ding, Zixiang and Xia, Rui and Yu, Jianfei , editor =. End-to-End Emotion-Cause Pair Extraction Based on Sliding Window Multi-Label Learning , booktitle =. 2020 , month = nov, pages =. doi:10.18653/v1/2020.emnlp-main.290 , url =

    doi:10.18653/v1/2020.emnlp-main.290 2020

  7. [7]

    Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics , author =

    ECPE-2D: Emotion-Cause Pair Extraction Based on Joint Two-Dimensional Representation, Interaction and Prediction , shorttitle =. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics , author =. 2020 , month = jul, pages =. doi:10.18653/v1/2020.acl-main.288 , url =

    doi:10.18653/v1/2020.acl-main.288 2020

  8. [8]

    2021 , journal =

    Multi-Task Sequence Tagging for Emotion-Cause Pair Extraction Via Tag Distribution Refinement , author =. 2021 , journal =. doi:10.1109/TASLP.2021.3089837 , url =

    arXiv 2021

  9. [9]

    Event-Driven Emotion Cause Extraction with Corpus Construction , booktitle =

    Gui, Lin and Wu, Dongyin and Xu, Ruifeng and Lu, Qin and Zhou, Yu , editor =. Event-Driven Emotion Cause Extraction with Corpus Construction , booktitle =. 2016 , month = nov, pages =. doi:10.18653/v1/D16-1170 , url =

    doi:10.18653/v1/d16-1170 2016

  10. [10]

    A Text-Driven Rule-Based System for Emotion Cause Detection , booktitle =

    Lee, Sophia Yat Mei and Chen, Ying and Huang, Chu-Ren , editor =. A Text-Driven Rule-Based System for Emotion Cause Detection , booktitle =. 2010 , month = jun, pages =

    2010

  11. [11]

    Towards Deep Conversational Recommendations , booktitle =

    Li, Raymond and Ebrahimi Kahou, Samira and Schulz, Hannes and Michalski, Vincent and Charlin, Laurent and Pal, Chris , year =. Towards Deep Conversational Recommendations , booktitle =

  12. [13]

    Findings of the Association for Computational Linguistics: ACL 2024 , author =

    LLM-REDIAL: A Large-Scale Dataset for Conversational Recommender Systems Created from User Behaviors with LLMs , shorttitle =. Findings of the Association for Computational Linguistics: ACL 2024 , author =. 2024 , month = aug, pages =. doi:10.18653/v1/2024.findings-acl.529 , url =

    doi:10.18653/v1/2024.findings-acl.529 2024

  13. [14]

    2022 , publisher =

    Sentiment Analysis and Opinion Mining , author =. 2022 , publisher =

    2022

  14. [15]

    Proceedings of the 10th International Workshop on Semantic Evaluation (SemEval-2016) , author =

    SemEval-2016 Task 5: Aspect Based Sentiment Analysis , shorttitle =. Proceedings of the 10th International Workshop on Semantic Evaluation (SemEval-2016) , author =. 2016 , month = jun, pages =. doi:10.18653/v1/S16-1002 , url =

    doi:10.18653/v1/s16-1002 2016

  15. [16]

    Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics , author =

    Towards Empathetic Open-Domain Conversation Models: A New Benchmark and Dataset , shorttitle =. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics , author =. 2019 , month = jul, pages =. doi:10.18653/v1/P19-1534 , url =

    doi:10.18653/v1/p19-1534 2019

  16. [18]

    Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics , author =

    Emotion-Cause Pair Extraction: A New Task to Emotion Analysis in Texts , shorttitle =. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics , author =. 2019 , month = jul, pages =. doi:10.18653/v1/P19-1096 , url =

    doi:10.18653/v1/p19-1096 2019

  17. [19]

    Towards Persona-Based Empathetic Conversational Models , booktitle =

    Zhong, Peixiang and Zhang, Chen and Wang, Hao and Liu, Yong and Miao, Chunyan , editor =. Towards Persona-Based Empathetic Conversational Models , booktitle =. 2020 , month = nov, pages =. doi:10.18653/v1/2020.emnlp-main.531 , url =

    doi:10.18653/v1/2020.emnlp-main.531 2020

  18. [20]

    2024 , month = feb, journal =

    A Knowledge-Guided Graph Attention Network for Emotion-Cause Pair Extraction , author =. 2024 , month = feb, journal =. doi:10.1016/j.knosys.2023.111342 , url =

    arXiv 2024

  19. [21]

    Transactions of the association for computational linguistics , volume=

    Spanbert: Improving pre-training by representing and predicting spans , author=. Transactions of the association for computational linguistics , volume=. 2020 , publisher=

    2020

  20. [22]

    2019 , eprint=

    RoBERTa: A Robustly Optimized BERT Pretraining Approach , author=. 2019 , eprint=

    2019

  21. [23]

    arXiv preprint arXiv:1711.05101 , year=

    Decoupled weight decay regularization , author=. arXiv preprint arXiv:1711.05101 , year=

    Pith/arXiv arXiv

  22. [24]

    PyTorch: an imperative style, high-performance deep learning library , year =

    Paszke, Adam and Gross, Sam and Massa, Francisco and Lerer, Adam and Bradbury, James and Chanan, Gregory and Killeen, Trevor and Lin, Zeming and Gimelshein, Natalia and Antiga, Luca and Desmaison, Alban and K\". PyTorch: an imperative style, high-performance deep learning library , year =. Proceedings of the 33rd International Conference on Neural Informa...

  23. [25]

    2018 , eprint=

    Graph Attention Networks , author=. 2018 , eprint=

    2018

  24. [26]

    International Conference on Learning Representations , year=

    How Attentive are Graph Attention Networks? , author=. International Conference on Learning Representations , year=

  25. [27]

    Proceedings of the 27th International Conference on Neural Information Processing Systems - Volume 2 , pages =

    Cuturi, Marco , title =. Proceedings of the 27th International Conference on Neural Information Processing Systems - Volume 2 , pages =. 2013 , address =

    2013

  26. [28]

    Proceedings of the 36th International Conference on Machine Learning , pages =

    Optimal Transport for structured data with application on graphs , author =. Proceedings of the 36th International Conference on Machine Learning , pages =. 2019 , editor =

    2019

  27. [29]

    PWEITINLP at S em E val-2024 Task 3: Two Step Emotion Cause Analysis

    Levchenko, Sofiia and Wolert, Rafa and Andruszkiewicz, Piotr. PWEITINLP at S em E val-2024 Task 3: Two Step Emotion Cause Analysis. Proceedings of the 18th International Workshop on Semantic Evaluation (SemEval-2024). 2024. doi:10.18653/v1/2024.semeval-1.159

    work page doi:10.18653/v1/2024.semeval-1.159 2024

  28. [30]

    P et K az at S em E val-2024 Task 3: Advancing Emotion Classification with an LLM for Emotion-Cause Pair Extraction in Conversations

    Kazakov, Roman and Petukhova, Kseniia and Kochmar, Ekaterina. P et K az at S em E val-2024 Task 3: Advancing Emotion Classification with an LLM for Emotion-Cause Pair Extraction in Conversations. Proceedings of the 18th International Workshop on Semantic Evaluation (SemEval-2024). 2024. doi:10.18653/v1/2024.semeval-1.164

    work page doi:10.18653/v1/2024.semeval-1.164 2024

  29. [31]

    JMI at S em E val 2024 Task 3: Two-step approach for multimodal ECAC using in-context learning with GPT and instruction-tuned Llama models

    Arefa and Ansari, Mohammed Abbas and Saxena, Chandni and Ahmad, Tanvir. JMI at S em E val 2024 Task 3: Two-step approach for multimodal ECAC using in-context learning with GPT and instruction-tuned Llama models. Proceedings of the 18th International Workshop on Semantic Evaluation (SemEval-2024). 2024. doi:10.18653/v1/2024.semeval-1.223

    work page doi:10.18653/v1/2024.semeval-1.223 2024

  30. [32]

    MIPS at S em E val-2024 Task 3: Multimodal Emotion-Cause Pair Extraction in Conversations with Multimodal Language Models

    Cheng, Zebang and Niu, Fuqiang and Lin, Yuxiang and Cheng, Zhi-qi and Peng, Xiaojiang and Zhang, Bowen. MIPS at S em E val-2024 Task 3: Multimodal Emotion-Cause Pair Extraction in Conversations with Multimodal Language Models. Proceedings of the 18th International Workshop on Semantic Evaluation (SemEval-2024). 2024. doi:10.18653/v1/2024.semeval-1.97

    work page doi:10.18653/v1/2024.semeval-1.97 2024

  31. [33]

    NUS -Emo at S em E val-2024 Task 3: Instruction-Tuning LLM for Multimodal Emotion-Cause Analysis in Conversations

    Luo, Meng and Zhang, Han and Wu, Shengqiong and Li, Bobo and Han, Hong and Fei, Hao. NUS -Emo at S em E val-2024 Task 3: Instruction-Tuning LLM for Multimodal Emotion-Cause Analysis in Conversations. Proceedings of the 18th International Workshop on Semantic Evaluation (SemEval-2024). 2024. doi:10.18653/v1/2024.semeval-1.226

    doi:10.18653/v1/2024.semeval-1.226 2024

  32. [34]

    U ni MEEC : Towards Unified Multimodal Emotion Recognition and Emotion Cause

    Hu, Guimin and Zhu, Zhihong and Hershcovich, Daniel and Hu, Lijie and Seifi, Hasti and Xie, Jiayuan. U ni MEEC : Towards Unified Multimodal Emotion Recognition and Emotion Cause. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024. doi:10.18653/v1/2024.findings-emnlp.302

    work page doi:10.18653/v1/2024.findings-emnlp.302 2024

  33. [35]

    An adaptive hybrid machine reading comprehension framework for multimodal emotion-cause pair extraction in conversations , journal =

    Guorui Li and Xufeng Duan and Cong Wang and Sancheng Peng , keywords =. An adaptive hybrid machine reading comprehension framework for multimodal emotion-cause pair extraction in conversations , journal =. 2025 , issn =. doi:https://doi.org/10.1016/j.engappai.2025.112333 , url =

    arXiv 2025

  34. [36]

    Proceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval , author =

    HCPRA: A Hierarchical Cognition–Perception–Reasoning Agent Framework for Emotion-Cause Pair Extraction in Conversations , shorttitle =. Proceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval , author =. 2026 , pages =. doi:10.1145/3805712.3809665 , eventtitle =

    arXiv 2026

  35. [37]

    Edward J Hu and Yelong Shen and Phillip Wallis and Zeyuan Allen-Zhu and Yuanzhi Li and Shean Wang and Lu Wang and Weizhu Chen , booktitle=. Lo

  36. [38]

    2024 , eprint=

    The Llama 3 Herd of Models , author =. 2024 , eprint=

    2024

  37. [39]

    2025 , eprint=

    Qwen2.5 Technical Report , author=. 2025 , eprint=

    2025

  38. [40]

    2024 , eprint=

    ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools , author=. 2024 , eprint=

    2024

  39. [41]

    Towards Reasoning in Large Language Models: A Survey

    Huang, Jie and Chang, Kevin Chen-Chuan. Towards Reasoning in Large Language Models: A Survey. Findings of the Association for Computational Linguistics: ACL 2023. 2023. doi:10.18653/v1/2023.findings-acl.67

    doi:10.18653/v1/2023.findings-acl.67 2023

  40. [42]

    Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence,

    Beyond the Limits: A Survey of Techniques to Extend the Context Length in Large Language Models , author =. Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence,. 2024 , month =. doi:10.24963/ijcai.2024/917 , url =

    doi:10.24963/ijcai.2024/917 2024

  41. [43]

    Computational Linguistics , volume=

    Large language model instruction following: A survey of progresses and challenges , author=. Computational Linguistics , volume=

  42. [44]

    2021 IEEE International Conference on Multimedia and Expo (ICME) , author =

    CONSK-GCN: Conversational Semantic- and Knowledge-Oriented Graph Convolutional Network for Multimodal Emotion Recognition , shorttitle =. 2021 IEEE International Conference on Multimedia and Expo (ICME) , author =. 2021 , month = jul, pages =. doi:10.1109/ICME51207.2021.9428438 , urldate =

    Pith/arXiv arXiv 2021

  43. [45]

    2022 , month = jul, journal =

    Context- and Knowledge-Aware Graph Convolutional Network for Multimodal Emotion Recognition , author =. 2022 , month = jul, journal =. doi:10.1109/MMUL.2022.3173430 , urldate =

    arXiv 2022

  44. [46]

    2023 , journal =

    Emotion Recognition in Conversations: A Survey Focusing on Context, Speaker Dependencies, and Fusion Methods , author =. 2023 , journal =

    2023

  45. [47]

    INTERNATIONAL CONFERENCE ON MULTIMODAL INTERACTION , author =

    GCFormer: A Graph Convolutional Transformer for Speech Emotion Recognition , shorttitle =. INTERNATIONAL CONFERENCE ON MULTIMODAL INTERACTION , author =. 2023 , month = oct, pages =. doi:10.1145/3577190.3614177 , urldate =

    arXiv 2023

  46. [48]

    Proceedings of the 14th Workshop on Computational Approaches to Subjectivity, Sentiment, & Social Media Analysis , author =

    SEC: Context-Aware Metric Learning for Efficient Emotion Recognition in Conversation , shorttitle =. Proceedings of the 14th Workshop on Computational Approaches to Subjectivity, Sentiment, & Social Media Analysis , author =. 2024 , month = aug, pages =

    2024

  47. [49]

    DialogueGCN: A Graph Convolutional Neural Network for Emotion Recognition in Conversation , shorttitle =. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) , author =. 2019 , month = nov, pages =. doi:10.18653/v1/D19-1015 , urldate =

    doi:10.18653/v1/d19-1015 2019

  48. [50]

    Benchmarking Cyber Harassment Dialogue Comprehension through Emotion-Informed Manifestations-Determinants Demarcation , booktitle =

    Ghosh, Soumitra and Singh, Gopendra Vikram and Arora, Jashn and Ekbal, Asif , year =. Benchmarking Cyber Harassment Dialogue Comprehension through Emotion-Informed Manifestations-Determinants Demarcation , booktitle =. doi:10.1609/aaai.v38i20.30207 , urldate =

    doi:10.1609/aaai.v38i20.30207

  49. [51]

    MMGCN: Multimodal Fusion via Deep Graph Convolution Network for Emotion Recognition in Conversation , shorttitle =. 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) , author =. 2021 , month = aug, pages =. doi:10.18653/...

    doi:10.18653/v1/2021.acl-long.440 2021

  50. [52]

    ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , author =

    MM-DFN: Multimodal Dynamic Fusion Network for Emotion Recognition in Conversations , shorttitle =. ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , author =. 2022 , month = may, pages =. doi:10.1109/ICASSP43922.2022.9747397 , urldate =

    arXiv 2022

  51. [53]

    D aily D ialog: A Manually Labelled Multi-turn Dialogue Dataset

    Li, Yanran and Su, Hui and Shen, Xiaoyu and Li, Wenjie and Cao, Ziqiang and Niu, Shuzi. D aily D ialog: A Manually Labelled Multi-turn Dialogue Dataset. Proceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2017

    2017

  52. [54]

    2023 , journal =

    GraphMFT: A Graph Network Based Multimodal Fusion Technique for Emotion Recognition in Conversation , author =. 2023 , journal =

    2023

  53. [55]

    Proceedings of the AAAI Conference on Artificial Intelligence , author =

    DialogueRNN: An Attentive RNN for Emotion Detection in Conversations , shorttitle =. Proceedings of the AAAI Conference on Artificial Intelligence , author =. 2019 , month = jul, volume =. doi:10.1609/aaai.v33i01.33016818 , urldate =

    doi:10.1609/aaai.v33i01.33016818 2019

  54. [56]

    Findings of the Association for Computational Linguistics: EMNLP 2021 , author =

    DialogueTRM: Exploring Multi-Modal Emotional Dynamics in a Conversation , shorttitle =. Findings of the Association for Computational Linguistics: EMNLP 2021 , author =. 2021 , month = nov, pages =. doi:10.18653/v1/2021.findings-emnlp.229 , urldate =

    doi:10.18653/v1/2021.findings-emnlp.229 2021

  55. [57]

    2024 , month = may, number =

    Revisiting Multimodal Emotion Recognition in Conversation from the Perspective of Graph Spectrum , author =. 2024 , month = may, number =. 2404.17862 , publisher =

    Pith/arXiv arXiv 2024

  56. [58]

    Proceedings of the 32nd ACM International Conference on Multimedia , author =

    Ada2I: Enhancing Modality Balance for Multimodal Conversational Emotion Recognition , shorttitle =. Proceedings of the 32nd ACM International Conference on Multimedia , author =. 2024 , month = oct, pages =. doi:10.1145/3664647.3681648 , urldate =

    arXiv 2024

  57. [59]

    2024 , month = jul, journal =

    Multi-Party Conversation Modeling for Emotion Recognition , author =. 2024 , month = jul, journal =. doi:10.1109/TAFFC.2023.3273589 , urldate =

    arXiv 2024

  58. [60]

    Directed Acyclic Graph Network for Conversational Emotion Recognition , booktitle =

    Shen, Weizhou and Wu, Siyue and Yang, Yunyi and Quan, Xiaojun , editor =. Directed Acyclic Graph Network for Conversational Emotion Recognition , booktitle =. 2021 , month = aug, pages =. doi:10.18653/v1/2021.acl-long.123 , urldate =

    doi:10.18653/v1/2021.acl-long.123 2021

  59. [61]

    2024 , month = aug, number =

    Efficient Long-Distance Latent Relation-Aware Graph Neural Network for Multi-Modal Emotion Recognition in Conversations , author =. 2024 , month = aug, number =. 2407.00119 , publisher =

    Pith/arXiv arXiv 2024

  60. [62]

    and Lee, Sungbok and Narayanan, Shrikanth S

    Busso, Carlos and Bulut, Murtaza and Lee, Chi-Chun and Kazemzadeh, Abe and Mower, Emily and Kim, Samuel and Chang, Jeannette N. and Lee, Sungbok and Narayanan, Shrikanth S. , year =. IEMOCAP: Interactive Emotional Dyadic Motion Capture Database , shorttitle =. Language Resources and Evaluation , volume =. doi:10.1007/s10579-008-9076-6 , urldate =

    doi:10.1007/s10579-008-9076-6

  61. [63]

    Proceedings of the AAAI Conference on Artificial Intelligence , author =

    Well, Now We Know! Unveiling Sarcasm: Initiating and Exploring Multimodal Conversations with Reasoning , shorttitle =. Proceedings of the AAAI Conference on Artificial Intelligence , author =. 2024 , month = mar, volume =. doi:10.1609/aaai.v38i17.29864 , urldate =

    doi:10.1609/aaai.v38i17.29864 2024

  62. [64]

    2023 , journal =

    SUNET: Speaker-Utterance Interaction Graph Neural Network for Emotion Recognition in Conversations , author =. 2023 , journal =

    2023

  63. [65]

    Emotion Recognition in Conversation via Dynamic Personality , booktitle =

    Wang, Yan and Wang, Bo and Zhao, Yachao and Zhao, Dongming and Jin, Xiaojia and Zhang, Jijun and He, Ruifang and Hou, Yuexian , editor =. Emotion Recognition in Conversation via Dynamic Personality , booktitle =. 2024 , month = may, pages =

    2024

  64. [66]

    AIMDiT: Modality Augmentation and Interaction via Multimodal Dimension Transformation for Emotion Recognition in Conversations , shorttitle =

    Wu, Sheng and Liu, Jiaxing and Wang, Longbiao and He, Dongxiao and Wang, Xiaobao and Dang, Jianwu , year =. AIMDiT: Modality Augmentation and Interaction via Multimodal Dimension Transformation for Emotion Recognition in Conversations , shorttitle =. 2407.00743 , publisher =

    Pith/arXiv arXiv

  65. [67]

    Emotion-Anchored Contrastive Learning Framework for Emotion Recognition in Conversation , booktitle =

    Yu, Fangxu and Guo, Junjie and Wu, Zhen and Dai, Xinyu , editor =. Emotion-Anchored Contrastive Learning Framework for Emotion Recognition in Conversation , booktitle =. 2024 , month = jun, pages =. doi:10.18653/v1/2024.findings-naacl.282 , urldate =

    doi:10.18653/v1/2024.findings-naacl.282 2024

  66. [68]

    Tracing Intricate Cues in Dialogue: Joint Graph Structure and Sentiment Dynamics for Multimodal Emotion Recognition , shorttitle =

    Li, Jiang and Wang, Xiaoping and Zeng, Zhigang , year =. Tracing Intricate Cues in Dialogue: Joint Graph Structure and Sentiment Dynamics for Multimodal Emotion Recognition , shorttitle =. IEEE Transactions on Pattern Analysis and Machine Intelligence , shortjournal =. doi:10.1109/TPAMI.2025.3581236 , langid =

    arXiv 2025

  67. [69]

    Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC'10) , month = may, year =

    Twitter as a Corpus for Sentiment Analysis and Opinion Mining , author =. Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC'10) , month = may, year =

  68. [70]

    2023 , journal =

    Global-View and Speaker-Aware Emotion Cause Extraction in Conversations , author =. 2023 , journal =. doi:10.1109/TASLP.2023.3319990 , urldate =

    arXiv 2023

  69. [71]

    SenticNet 5: Discovering Conceptual Primitives for Sentiment Analysis by Means of Context Embeddings , shorttitle =

    Cambria, Erik and Poria, Soujanya and Hazarika, Devamanyu and Kwok, Kenneth , year =. SenticNet 5: Discovering Conceptual Primitives for Sentiment Analysis by Means of Context Embeddings , shorttitle =. Proceedings of the AAAI Conference on Artificial Intelligence , volume =. doi:10.1609/aaai.v32i1.11559 , urldate =

    doi:10.1609/aaai.v32i1.11559

  70. [72]

    Conversational Emotion-Cause Pair Extraction with Guided Mixture of Experts , booktitle =

    Jeong, DongJin and Bak, JinYeong , editor =. Conversational Emotion-Cause Pair Extraction with Guided Mixture of Experts , booktitle =. 2023 , month = may, pages =. doi:10.18653/v1/2023.eacl-main.240 , urldate =

    doi:10.18653/v1/2023.eacl-main.240 2023

  71. [73]

    2022 , month = oct, number =

    Discourse-Aware Emotion Cause Extraction in Conversations , author =. 2022 , month = oct, number =. doi:10.48550/arXiv.2210.14419 , urldate =. 2210.14419 , primaryclass =

    Pith/arXiv arXiv doi:10.48550/arxiv.2210.14419 2022

  72. [74]

    Enhancing Emotion--Cause Pair Extraction in Conversation With Contextual Information , author =

  73. [75]

    ECPEC: Emotion-Cause Pair Extraction in Conversations , shorttitle =

    Li, Wei and Li, Yang and Pandelea, Vlad and Ge, Mengshi and Zhu, Luyao and Cambria, Erik , year =. ECPEC: Emotion-Cause Pair Extraction in Conversations , shorttitle =. IEEE Transactions on Affective Computing , volume =. doi:10.1109/TAFFC.2022.3216551 , urldate =

    arXiv 2022

  74. [76]

    2024 , journal =

    Multimodal Emotion-Cause Pair Extraction with Holistic Interaction and Label Constraint , author =. 2024 , journal =. doi:10.1145/3689646 , urldate =

    doi:10.1145/3689646 2024

  75. [77]

    Emotion-Cause Pair Extraction in Conversations Based on Multi-Turn MRC with Position-Aware GCN , booktitle =

    Liu, Chen and Niu, Changyong and Xie, Jinge and Jia, Yuxiang and Zan, Hongying , year =. Emotion-Cause Pair Extraction in Conversations Based on Multi-Turn MRC with Position-Aware GCN , booktitle =. doi:10.1109/IALP61005.2023.10337182 , urldate =

    Pith/arXiv arXiv 2023

  76. [78]

    2021 , month = sep, journal =

    Recognizing Emotion Cause in Conversations , author =. 2021 , month = sep, journal =. doi:10.1007/s12559-021-09925-7 , urldate =

    doi:10.1007/s12559-021-09925-7 2021

  77. [79]

    2023 , journal =

    Recent Trends in Deep Learning Based Textual Emotion Cause Extraction , author =. 2023 , journal =. doi:10.1109/TASLP.2023.3254166 , urldate =

    arXiv 2023

  78. [80]

    2023 , month = jul, journal =

    Multimodal Emotion-Cause Pair Extraction in Conversations , author =. 2023 , month = jul, journal =. doi:10.1109/TAFFC.2022.3226559 , urldate =

    arXiv 2023

  79. [81]

    Enhancing Emotion-Cause Pair Extraction in Conversations via Center Event Detection and Reasoning , booktitle =

    Wang, Botao and Tang, Keke and Zhu, Peican , editor =. Enhancing Emotion-Cause Pair Extraction in Conversations via Center Event Detection and Reasoning , booktitle =. 2024 , month = nov, pages =. doi:10.18653/v1/2024.findings-emnlp.632 , urldate =

    doi:10.18653/v1/2024.findings-emnlp.632 2024

  80. [82]

    Scientific Reports , volume =

    MULTICAUSENET temporal attention for multimodal emotion cause pair extraction , author =. Scientific Reports , volume =. 2025 , doi =

    2025

Showing first 80 references.

This paper was first reviewed by grok-4.5 on July 30, 2026.