Pith. sign in

REVIEW 4 major objections 5 minor 93 references

Dialogue-level affective atmosphere is a reusable prior that steers utterance emotion prediction better than raw global context.

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-4.5

2026-07-30 22:52 UTC pith:HVSKG5VZ

load-bearing objection Solid ERC methods paper with a real dual-use prior idea; the continuous graph prior helps the lightweight path, but the LLM “atmosphere” plug-in is mostly majority-emotion label injection. the 4 major comments →

arxiv 2607.26726 v1 pith:HVSKG5VZ submitted 2026-07-29 cs.CL

AtmosERC: Modeling Dialogue-Level Affective Atmosphere for Emotion Recognition in Conversation

Weijie Feng , Tongwei Zhang , Binbin Liu , Zhiyong Cheng This is my paper
classification cs.CL
keywords emotion recognition in conversationaffective atmosphereconversational graphrelation-aware GNNaffective priorLLM promptingspeaker modeling
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.

Emotion recognition in conversation usually treats the whole dialogue context as one mixed signal. This paper argues that only the affect-oriented part of that signal—the dialogue’s latent affective atmosphere—should be isolated and reused. AtmosERC builds a conversational graph over utterances and speakers, filters relation-specific signals, and compresses them into a compact atmosphere prior plus speaker-conditioned priors. That prior initializes a lightweight sequential decoder and can be verbalized as a short prompt cue for large language models without changing their weights. On four standard benchmarks the method improves lightweight models, lifts several LLM baselines as a plug-in, and helps predictions recover after local emotional deviations, especially when the dialogue has a clear dominant affective tendency.

Core claim

Dialogue-level affective atmosphere—an affect-oriented abstraction of global context estimated by relation-aware graph filtering over utterance and speaker nodes—acts as an explicit, decoder-independent prior that improves lightweight ERC, enhances LLM-based ERC as an unmodified prompt plug-in, and stabilizes predictions after local emotional deviations.

What carries the argument

Relation-aware graph atmosphere extractor: a heterogeneous conversational graph (inter-/intra-speaker edges, semantic similarity, utterance–speaker affiliation) whose layered relation-specific GNN filtering and max-pooling yield a compact dialogue atmosphere vector a and speaker-conditioned priors that initialize decoding or are verbalized for prompts.

Load-bearing premise

That max-pooled graph states from frozen text features, trained partly to match a dialogue’s dominant emotion, truly isolate latent atmosphere rather than a generic context or majority-label summary—especially in short or highly neutral dialogues.

What would settle it

Replace the learned atmosphere prior with plain average pooling of the same utterance features (or inject a wrong dominant-emotion cue) and check whether the reported gains on IEMOCAP/MELD and the local-deviation recovery scores disappear.

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

If this is right

  • Lightweight ERC can stay competitive by conditioning a small sequential decoder on an explicit atmosphere prior instead of heavier end-to-end context encoders.
  • LLM-based ERC can gain from a short plug-in atmosphere phrase without fine-tuning or changing the backbone prompt template.
  • Models guided by atmosphere should recover the dominant trajectory after a single off-atmosphere utterance more often than pure local-context baselines.
  • When dialogues are short or mostly neutral, atmosphere cues weaken and may need richer supervision or multimodal signals.

Where Pith is reading between the lines

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

  • Atmosphere estimation quality is the binding constraint: gold dominant-emotion proxies already beat the learned verbalizer, so better atmosphere heads could unlock larger LLM gains.
  • The same compact prior could be shared across related dialogue tasks (empathy response, tone-controlled generation) as a cheap global affective control signal.
  • Failure modes on neutral-heavy data suggest atmosphere may need continuous valence/arousal targets rather than discrete majority emotion alone.

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

4 major / 5 minor

Summary. The paper argues that global conversational context is heterogeneous and that an affect-oriented component—dialogue-level affective atmosphere—can be estimated and reused for Emotion Recognition in Conversation (ERC). AtmosERC builds a heterogeneous conversational graph over utterance and speaker nodes with four hand-specified relations, applies relation-specific GNN filtering, and max-pools filtered utterance states into a compact prior a (with speaker-conditioned rows H(s) as auxiliaries). This prior initializes a lightweight BiGRU/speaker-state decoder and, after a verbalization head trained to predict the dialogue’s dominant emotion, is injected as a single emotion-name cue into frozen LLM ERC prompts. On IEMOCAP, MELD, EmoryNLP, and DailyDialog the method reports gains for lightweight ERC, mixed paired gains for LLM plug-in, improved recovery after local emotional deviations, and ablations plus appendix diagnostics against mean-pooled RoBERTa context.

Significance. If atmosphere is genuinely an affect-filtered reusable prior rather than a majority-label or generic context summary, the dual-use design (lightweight decoder guidance plus backbone-agnostic LLM cue) is a useful contribution to ERC: it separates global affective tendency from utterance decoding and offers a practical plug-in path for LLM systems without fine-tuning. Strengths include multi-benchmark evaluation, component/relation ablations (Table 5), paired LLM comparisons (Table 3), local-deviation recovery (Table 4), and explicit prior-vs-context checks (Appendix D.3, Tables 9–10). The work is incremental relative to graph ERC and context pooling, but the framing and dual interface are clear enough to matter if the isolation claim is tightened.

major comments (4)
  1. [§3.4, Eqs. (9)–(10); Appendix D.2] §3.4, Eqs. (9)–(10): the LLM plug-in does not consume the continuous prior a. A head is trained so that verbalization predicts the dialogue dominant emotion z★ and only that single emotion name is prepended. The dual-use claim that atmosphere is a reusable affect-oriented prior for unmodified LLM ERC therefore reduces, at the interface, to majority-label injection. Appendix D.2’s larger gold-proxy gains reinforce that the prompt channel is majority-collapsed. Either inject a richer discretization of a (e.g., multi-label/soft/valence descriptors, or several latent clusters independent of z★), or narrow the abstract/claims so that LLM enhancement is explicitly “dominant-emotion cue estimated from the graph prior,” not atmosphere as a latent continuous prior.
  2. [§3.1; Fig. 2; Appendix D.3; §5.1–5.2] Motivation and diagnostics lean on the same dominant-emotion proxy used for verbalization (Fig. 2a; §3.1; Appendix D.3 distance analysis with z★). Lightweight gains vs mean-pooled context c (Table 9) and strength buckets (Table 10) are the main evidence that a is not generic context, but they do not show that a is not largely a soft majority-emotion embedding. A load-bearing check is missing: e.g., linear probe of a for dominant emotion vs topic/length/speaker-count controls; conditional mutual information; or performance when majority labels are shuffled while local utterance labels are fixed. Without this, the central “affect-oriented atmosphere” interpretation remains under-supported relative to a majority-summary reading—especially where the paper already reports weak/negative results (DailyDialog neutrality, §5.1; LaERC-S on MELD −0.57, §5.2).
  3. [Abstract; Table 2; Table 3; §5.1–5.2] Table 2 / §5.1: AtmosERC trails SKIER by 2.63 micro-F1 on DailyDialog while leading elsewhere; Table 3 shows a negative plug-in result for LaERC-S on MELD. The abstract’s unqualified claim that the method “improves lightweight ERC” and “enhances LLM-based ERC as a plug-in cue” overstates uniformity. Please qualify claims by dialogue length/neutrality regime, and add failure analysis (short multi-party shifts; 83% neutral DailyDialog) so the scope of the atmosphere prior is falsifiable rather than averaged away.
  4. [§3.2–3.3; Table 5] §3.2–3.3 and Table 5: the lightweight decoder is atmosphere-initialized BiGRU + speaker GRU on frozen RoBERTa features. Gains vs DAG-ERC and “w/o GAE” (non-structured pooling) are informative, but it remains unclear how much comes from the specific max-pool atmosphere bottleneck versus simply adding a strong relational encoder plus speaker states. A control that uses the same graph encoder to refine utterance nodes for direct classification (standard graph ERC head) without collapsing to a single dialogue vector a would test whether the atmosphere bottleneck itself is necessary for the reported lightweight gains.
minor comments (5)
  1. [Figure 1; Appendix D.3] Figure 1 caption refers to “atmosphere distance” while Appendix D.3 clarifies the plotted quantity is often a dominant-emotion proxy distance; align main-text figure caption and axis labels with D.3 to avoid implying supervised atmosphere labels exist.
  2. [Table 2] Table 2 leaves many baseline cells as “-” or “–”; state explicitly whether missing numbers mean unavailable under the same text-only protocol or not re-run, to aid fair comparison.
  3. [Table 6; Appendix C.3] Hyperparameters (Table 6) vary W, τs, L, dropout substantially by dataset; a short sensitivity plot for τs and W would help assess robustness of the four-relation design.
  4. [§3.3; References] Typos/notation: “V ossen” spacing in citations; “centered on two questions” → “centers”; BiGRUu(x; a) initialization is described in prose but not notated in Eq. (6).
  5. [§4.3] Code/data “will be released upon acceptance”—for reproducibility of graph construction and paired LLM prompts, a minimal anonymous release or config dump at review would strengthen confidence.

Circularity Check

2 steps flagged

Continuous atmosphere prior is independently graph-extracted; only the LLM verbalization path collapses ‘atmosphere’ to a dominant-emotion label by training target.

specific steps
  1. self definitional [§3.4 Eqs. 9–10; Atmosphere Verbalization]
    "Since atmosphere has no direct annotation, we first derive a dialogue-level proxy target from the dominant emotion in each training dialogue: z★ = arg max_y∈Y ∑_i I(yi=y). We then train a lightweight verbalization head to predict this proxy from the atmosphere vector a … bz is used as the textual descriptor for prompt injection. … The overall affective atmosphere of this dialogue is <ATMOSPHERE>."

    For the LLM path the paper does not inject the continuous prior a. It defines the only supervision for verbalization as the dialogue majority emotion z★ and trains FCv so bz ≈ z★. The prompt cue called ‘atmosphere’ is therefore the dominant-emotion estimate by construction of Eqs. 9–10, not an independent affect descriptor. LLM ‘atmosphere’ gains are gains from injecting (approx.) majority class; Appendix D.2’s larger gold-proxy gains make that reduction explicit. This definitional collapse affects the reusable-prior/LLM-plug-in subclaim only—not the graph extraction of a or lightweight decoding.

  2. fitted input called prediction [Appendix D.2 Table 8; cf. §5.2]
    "For each dialogue, the proxy descriptor is defined as the dominant emotion, i.e., the most frequent utterance-level emotion label in that dialogue. … replacing the predicted atmosphere descriptor with this proxy yields larger gains across all paired ERC-specific LLM comparisons. … the gap between predicted and proxy descriptors indicates remaining headroom in estimating and verbalizing atmosphere"

    The diagnostic treats better majority-label injection as an upper bound on atmosphere prompting. That frames success of AtmosERC-P as fidelity of a→z★, i.e., quality of a fitted dominant-emotion predictor, then reports it as headroom in ‘atmosphere’ estimation. The predicted quantity and the fit target are the same label family; the ‘prediction’ improvement is statistically the same objective used to train the verbalizer.

full rationale

AtmosERC’s core object a is not defined as the quantity it is later asked to explain. Utterance features come from frozen RoBERTa; a is max-pooled after relation-specific GNN filtering (Eqs. 1–5) and is used as continuous initialization for a BiGRU/speaker decoder trained with standard utterance-level cross-entropy. That path is ordinary representation learning against external ERC labels and does not reduce by construction to its inputs. Circularity pressure is localized to the LLM plug-in and to proxy-based diagnostics: with no atmosphere annotations, §3.4 defines the verbalization target z★ as the dialogue’s dominant emotion and trains FCv(a) so that the injected cue bz is exactly a predicted majority label; Appendix D.2’s gold-proxy upper bound then shows larger gains from the true majority label, confirming the interface is majority-collapsed. Motivation and recovery analyses likewise lean on dominant-emotion concentration (Fig. 2, §5.3). That is proxy supervision and construct narrowing for one subclaim, not a full self-referential derivation of the method. No load-bearing self-citation or uniqueness import appears. Score 3 reflects partial definitional collapse on the verbalized LLM cue only; lightweight gains and a-vs-context checks remain non-circular.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 2 invented entities

The central claim rests on standard ERC task setup and GNN practice, plus paper-specific modeling choices: that heterogeneous conversational relations can be filtered into a single affect prior; that max pooling and dominant-emotion verbalization are adequate atmosphere interfaces; and several dataset-tuned graph thresholds. No new physical entities; the invented construct is the atmosphere prior itself.

free parameters (5)
  • semantic similarity threshold τs = 0.95 IEMOCAP/MELD; 0.90 EmoryNLP; 0.70 DailyDialog
    Controls r3 long-range edges; selected on validation and set differently per dataset (0.95/0.95/0.90/0.70).
  • inter-speaker window W = 4 IEMOCAP; 1 others
    Defines r1 neighborhood size; chosen per dataset rather than derived.
  • GNN depth L and dropout = L=2/2/2/1; dropout 0.1/0.1/0.2/0.3
    Architecture regularization knobs tuned per dataset (L=2 or 1; dropout 0.1–0.3).
  • atmosphere embedding dimension d and training hyperparameters = d=1024; lr=5e-5; epochs 150/80/50/50
    Hidden size 1024, AdamW lr 5e-5, weight decay 1e-2, batch sizes and epoch counts set per dataset; standard but claim-relevant fit choices.
  • verbalization head mapping a → dominant-emotion descriptor = softmax(FCv(a)) trained on training dialogues
    Trained to predict z★ = argmax emotion count; discretizes continuous atmosphere for LLM prompts and ties the cue to majority labels.
axioms (6)
  • domain assumption ERC labels depend on utterance semantics, history, and speaker interactions; global context is a useful but heterogeneous signal.
    Standard ERC premise stated in §1 and used to motivate filtering context into atmosphere.
  • domain assumption GNNs act as low-pass filters and therefore favor stable dialogue-level affective tendency over transient turn noise.
    Invoked in §3.2 citing NT & Maehara 2019 as design justification for graph atmosphere extraction.
  • ad hoc to paper Four hand-specified relations (inter-speaker window, intra-speaker chain, cosine similarity edges, utterance–speaker affiliation) adequately span affective dependencies needed for atmosphere.
    Graph construction in §3.2; ablations support usefulness but the relation inventory is a modeling choice, not derived.
  • ad hoc to paper Dialogue atmosphere can be represented as one shared vector a obtained by max-pooling filtered utterance nodes, with speaker rows of H(s) as auxiliary priors.
    Eqs. 4–5 and §3.3 initialization scheme; alternative poolings are only partly probed via context mean baseline in appendix.
  • ad hoc to paper Dominant-emotion statistics are acceptable observable proxies for latent atmosphere when supervising verbalization or analyzing regularity.
    §3.1, Eq. 9, Fig. 2, App. D.2–D.3; load-bearing for LLM plug-in and for claiming atmosphere is affect-oriented.
  • domain assumption Frozen RoBERTa utterance features are a sufficient base representation for atmosphere extraction without updating the encoder.
    §3.1 textual feature extraction; common ERC practice but fixes the feature geometry the graph filters.
invented entities (2)
  • dialogue-level affective atmosphere prior a no independent evidence
    purpose: Compact affect-oriented global state guiding lightweight decoding and LLM prompts, separated from full context.
    Core postulated construct (§1, §3.1). Supported only by indirect proxies (dominant emotion, valence trajectories, context-distance scatter) and downstream ERC gains—no independent external measurement protocol.
  • speaker-conditioned affective priors H(s) no independent evidence
    purpose: Initialize per-speaker GRU states so local decoding is conditioned on speaker-level affective summary.
    Introduced in §3.2–3.3 via r4 aggregation; ablation w/o SSU hurts, but entity is internal to the architecture.

pith-pipeline@v1.2.0-daily-grok45 · 22571 in / 3980 out tokens · 78805 ms · 2026-07-30T22:52:27.437221+00:00 · methodology

0 comments
read the original abstract

Emotion Recognition in Conversation (ERC) aims to predict utterance-level emotions in dialogues and has largely advanced through context-centric modeling. However, global context is a heterogeneous signal, and not all contextual information is equally relevant to emotion prediction. This paper focuses on the affect-oriented component of this signal, termed dialogue-level affective atmosphere, which captures a latent tendency commonly reflected in conversational emotion patterns. To estimate and exploit this tendency, we propose AtmosERC, a graph-based ERC framework that models each dialogue as a conversational graph over utterances and speakers. A relation-aware graph extractor filters and fuses heterogeneous graph signals to produce dialogue-level and speaker-conditioned affective priors. The resulting compact prior guides lightweight sequential emotion prediction and can also be verbalized into prompt-level cues for LLM-based ERC without modifying backbone models. Experiments on four ERC benchmarks show that AtmosERC improves lightweight ERC, enhances LLM-based ERC as a plug-in cue, and yields more stable predictions under local emotional deviations.

Figures

Figures reproduced from arXiv: 2607.26726 by Binbin Liu, Tongwei Zhang, Weijie Feng, Zhiyong Cheng.

Figure 1
Figure 1. Figure 1: Relationship between global context distance and [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Empirical evidence for affective atmosphere. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overall architecture of our AtmosERC. 3 Methodology We present AtmosERC, an atmosphere-aware framework for ERC, as illustrated in [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Detailed visualization of global context distance [PITH_FULL_IMAGE:figures/full_fig_p012_4.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

93 extracted references · 3 linked inside Pith

  1. [1]

    Aho and Jeffrey D

    Alfred V. Aho and Jeffrey D. Ullman , title =. 1972 , volume =

    1972

  2. [2]

    Publications Manual , year =

  3. [3]

    Chandra and Dexter C

    Ashok K. Chandra and Dexter C. Kozen and Larry J. Stockmeyer , year =. Alternation , journal =

  4. [4]

    Scalable training of

    Andrew, Galen and Gao, Jianfeng , booktitle=. Scalable training of

  5. [5]

    1997 , publisher =

    Dan Gusfield , title =. 1997 , publisher =

    1997

  6. [6]

    Tetreault , title =

    Mohammad Sadegh Rasooli and Joel R. Tetreault , title =. Computing Research Repository , volume =. 2015 , url =

    2015

  7. [7]

    A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =

    Ando, Rie Kubota and Zhang, Tong , Issn =. A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =. Journal of Machine Learning Research , Numpages =

  8. [8]

    Correlation-Decoupled Knowledge Distillation for Multimodal Sentiment Analysis with Incomplete Modalities , booktitle =

    Li, Mingcheng and Yang, Dingkang and Zhao, Xiao and Wang, Shuaibing and Wang, Yan and Yang, Kun and Sun, Mingyang and Kou, Dongliang and Qian, Ziyun and Zhang, Lihua , year = 2024, pages =. Correlation-Decoupled Knowledge Distillation for Multimodal Sentiment Analysis with Incomplete Modalities , booktitle =

    2024

  9. [9]

    Scientific Reports , volume =

    A Multimodal Approach to Cross-Lingual Sentiment Analysis with Ensemble of Transformer and LLM , author =. Scientific Reports , volume =

  10. [10]

    International Journal of Data Science and Analytics , volume =

    A Review of Sentiment Analysis: Tasks, Applications, and Deep Learning Techniques , author =. International Journal of Data Science and Analytics , volume =

  11. [11]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume =

    TextGT: A Double-View Graph Transformer on Text for Aspect-Based Sentiment Analysis , author =. Proceedings of the AAAI Conference on Artificial Intelligence , volume =

  12. [12]

    Proceedings of the AAAI Conference on Artificial Intelligence , year =

    Ai, Wei and Zhang, Fuchen and Shou, Yuntao and Meng, Tao and Chen, Haowen and Li, Keqin , title =. Proceedings of the AAAI Conference on Artificial Intelligence , year =

  13. [13]

    Speaker-Guided Encoder-Decoder Framework for Emotion Recognition in Conversation , booktitle =

    Bao, Yinan and Ma, Qianwen and Wei, Lingwei and Zhou, Wei and Hu, Songlin , year = 2022, pages =. Speaker-Guided Encoder-Decoder Framework for Emotion Recognition in Conversation , booktitle =

    2022

  14. [14]

    Language Resources and Evaluation , volume =

    IEMOCAP: Interactive Emotional Dyadic Motion Capture Database , author =. Language Resources and Evaluation , volume =

  15. [15]

    Learning What and When to Drop: Adaptive Multimodal and Contextual Dynamics for Emotion Recognition in Conversation , booktitle =

    Chen, Feiyu and Sun, Zhengxiao and Ouyang, Deqiang and Liu, Xueliang and Shao, Jie , year = 2021, pages =. Learning What and When to Drop: Adaptive Multimodal and Contextual Dynamics for Emotion Recognition in Conversation , booktitle =

    2021

  16. [16]

    Multivariate, Multi-Frequency and Multimodal: Rethinking Graph Neural Networks for Emotion Recognition in Conversation , booktitle =

    Chen, Feiyu and Shao, Jie and Zhu, Shuyuan and Shen, Heng Tao , year = 2023, pages =. Multivariate, Multi-Frequency and Multimodal: Rethinking Graph Neural Networks for Emotion Recognition in Conversation , booktitle =

    2023

  17. [17]

    CONSK-GCN: Conversational Semantic- and Knowledge-Oriented Graph Convolutional Network for Multimodal Emotion Recognition , booktitle =

    Fu, Yahui and Okada, Shogo and Wang, Longbiao and Guo, Lili and Song, Yaodong and Liu, Jiaxing and Dang, Jianwu , year = 2021, pages =. CONSK-GCN: Conversational Semantic- and Knowledge-Oriented Graph Convolutional Network for Multimodal Emotion Recognition , booktitle =

    2021

  18. [18]

    IEEE MultiMedia , volume =

    Context- and Knowledge-Aware Graph Convolutional Network for Multimodal Emotion Recognition , author =. IEEE MultiMedia , volume =

  19. [19]

    Electronics , volume =

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

  20. [20]

    Proceedings of the 31st International Conference on Computational Linguistics

  21. [21]

    SEC: Context-Aware Metric Learning for Efficient Emotion Recognition in Conversation , booktitle =

    Gendron, Barbara and Guibon, Ga. SEC: Context-Aware Metric Learning for Efficient Emotion Recognition in Conversation , booktitle =

  22. [22]

    DialogueGCN: A Graph Convolutional Neural Network for Emotion Recognition in Conversation , booktitle =

    Ghosal, Deepanway and Majumder, Navonil and Poria, Soujanya and Chhaya, Niyati and Gelbukh, Alexander , year = 2019, pages =. DialogueGCN: A Graph Convolutional Neural Network for Emotion Recognition in Conversation , booktitle =

    2019

  23. [23]

    COSMIC: COmmonSense Knowledge for eMotion Identification in Conversations , booktitle =

    Ghosal, Deepanway and Majumder, Navonil and Gelbukh, Alexander and Mihalcea, Rada and Poria, Soujanya , year = 2020, pages =. COSMIC: COmmonSense Knowledge for eMotion Identification in Conversations , booktitle =

    2020

  24. [24]

    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 = 2024, volume =. Benchmarking Cyber Harassment Dialogue Comprehension through Emotion-Informed Manifestations-Determinants Demarcation , booktitle =

    2024

  25. [25]

    Conversational Memory Network for Emotion Recognition in Dyadic Dialogue Videos , booktitle =

    Hazarika, Devamanyu and Poria, Soujanya and Zadeh, Amir and Cambria, Erik and Morency, Louis-Philippe and Zimmermann, Roger , year = 2018, pages =. Conversational Memory Network for Emotion Recognition in Dyadic Dialogue Videos , booktitle =

    2018

  26. [26]

    MMGCN: Multimodal Fusion via Deep Graph Convolution Network for Emotion Recognition in Conversation , booktitle =

    Hu, Jingwen and Liu, Yuchen and Zhao, Jinming and Jin, Qin , year = 2021, pages =. MMGCN: Multimodal Fusion via Deep Graph Convolution Network for Emotion Recognition in Conversation , booktitle =

    2021

  27. [27]

    DialogueCRN: Contextual Reasoning Networks for Emotion Recognition in Conversations , booktitle =

    Hu, Dou and Wei, Lingwei and Huai, Xiaoyong , year = 2021, pages =. DialogueCRN: Contextual Reasoning Networks for Emotion Recognition in Conversations , booktitle =

    2021

  28. [28]

    MM-DFN: Multimodal Dynamic Fusion Network for Emotion Recognition in Conversations , booktitle =

    Hu, Dou and Hou, Xiaolong and Wei, Lingwei and Jiang, Lianxin and Mo, Yang , year = 2022, pages =. MM-DFN: Multimodal Dynamic Fusion Network for Emotion Recognition in Conversations , booktitle =

    2022

  29. [29]

    UniMSE: Towards Unified Multimodal Sentiment Analysis and Emotion Recognition , booktitle =

    Hu, Guimin and Lin, Ting-En and Zhao, Yi and Lu, Guangming and Wu, Yuchuan and Li, Yongbin , year = 2022, pages =. UniMSE: Towards Unified Multimodal Sentiment Analysis and Emotion Recognition , booktitle =

    2022

  30. [30]

    Supervised Adversarial Contrastive Learning for Emotion Recognition in Conversations , booktitle =

    Hu, Dou and Bao, Yinan and Wei, Lingwei and Zhou, Wei and Hu, Songlin , year = 2023, pages =. Supervised Adversarial Contrastive Learning for Emotion Recognition in Conversations , booktitle =

    2023

  31. [31]

    Beyond Emotion Recognition: A Multi-Turn Multimodal Emotion Understanding and Reasoning Benchmark , booktitle =

    Hu, Jinpeng and Shi, Hongchang and Dai, Chongyuan and Li, Zhuo and Song, Peipei and Wang, Meng , year = 2025, pages =. Beyond Emotion Recognition: A Multi-Turn Multimodal Emotion Understanding and Reasoning Benchmark , booktitle =

    2025

  32. [32]

    Exploiting Unsupervised Data for Emotion Recognition in Conversations , booktitle =

    Jiao, Wenxiang and Lyu, Michael and King, Irwin , editor =. Exploiting Unsupervised Data for Emotion Recognition in Conversations , booktitle =

  33. [33]

    Causal-ERC: A Multimodal Framework with Causal Prompting for Emotion Recognition in Conversations with Large Language Models , booktitle =

    Jing, Ran and. Causal-ERC: A Multimodal Framework with Causal Prompting for Emotion Recognition in Conversations with Large Language Models , booktitle =

  34. [34]

    Beyond Single Emotion: Multi-Label Approach to Conversational Emotion Recognition , booktitle =

    Kang, Yujin and Cho, Yoon-Sik , year = 2025, volume=. Beyond Single Emotion: Multi-Label Approach to Conversational Emotion Recognition , booktitle =

    2025

  35. [35]

    PFA-ERC: Psuedo-Future Augmented Dynamic Emotion Recognition in Conversations , booktitle =

    Khule, Tanmay and Agrawal, Rishabh and Narayan, Apurva , year = 2024, pages =. PFA-ERC: Psuedo-Future Augmented Dynamic Emotion Recognition in Conversations , booktitle =

    2024

  36. [36]

    2108.12009 , archivePrefix=

    EmoBERTa: Speaker-Aware Emotion Recognition in Conversation with RoBERTa , author =. 2108.12009 , archivePrefix=

    Pith/arXiv arXiv

  37. [37]

    2309.11911 , archivePrefix=

    InstructERC: Reforming Emotion Recognition in Conversation with Multi-Task Retrieval-Augmented Large Language Models , author =. 2309.11911 , archivePrefix=

    Pith/arXiv arXiv

  38. [38]

    DailyDialog: A Manually Labelled Multi-Turn Dialogue Dataset , booktitle =

    Li, Yanran and Su, Hui and Shen, Xiaoyu and Li, Wenjie and Cao, Ziqiang and Niu, Shuzi , year = 2017, pages =. DailyDialog: A Manually Labelled Multi-Turn Dialogue Dataset , booktitle =

    2017

  39. [39]

    HiTrans: A Transformer-Based Context- and Speaker-Sensitive Model for Emotion Detection in Conversations , booktitle =

    Li, Jingye and Ji, Donghong and Li, Fei and Zhang, Meishan and Liu, Yijiang , year = 2020, pages =. HiTrans: A Transformer-Based Context- and Speaker-Sensitive Model for Emotion Detection in Conversations , booktitle =

    2020

  40. [40]

    Past, Present, and Future: Conversational Emotion Recognition through Structural Modeling of Psychological Knowledge , booktitle =

    Li, Jiangnan and Lin, Zheng and Fu, Peng and Wang, Weiping , year = 2021, pages =. Past, Present, and Future: Conversational Emotion Recognition through Structural Modeling of Psychological Knowledge , booktitle =

    2021

  41. [41]

    Neurocomputing , volume =

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

  42. [42]

    Knowledge-Based Systems , volume =

    ERNetCL: A Novel Emotion Recognition Network in Textual Conversation Based on Curriculum Learning Strategy , author =. Knowledge-Based Systems , volume =

  43. [43]

    IEEE Transactions on Multimedia , volume =

    GraphCFC: A Directed Graph Based Cross-Modal Feature Complementation Approach for Multimodal Conversational Emotion Recognition , author =. IEEE Transactions on Multimedia , volume =

  44. [44]

    IEEE Transactions on Affective Computing , volume =

    CFN-ESA: A Cross-Modal Fusion Network With Emotion-Shift Awareness for Dialogue Emotion Recognition , author =. IEEE Transactions on Affective Computing , volume =

  45. [45]

    IEEE Transactions on Affective Computing , volume =

    GA2MIF: Graph and Attention Based Two-Stage Multi-Source Information Fusion for Conversational Emotion Detection , author =. IEEE Transactions on Affective Computing , volume =

  46. [46]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =

    Tracing Intricate Cues in Dialogue: Joint Graph Structure and Sentiment Dynamics for Multimodal Emotion Recognition , author =. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =

  47. [47]

    2508.01318 , archivePrefix=

    AffectGPT-R1: Leveraging Reinforcement Learning for Open-Vocabulary Multimodal Emotion Recognition , author =. 2508.01318 , archivePrefix=

    arXiv

  48. [48]

    IEEE Transactions on Multimedia , volume =

    A Transformer-Based Model With Self-Distillation for Multimodal Emotion Recognition in Conversations , author =. IEEE Transactions on Multimedia , volume =

  49. [49]

    In-Context Examples Matter: Improving Emotion Recognition in Conversation with Instruction Tuning , author =

  50. [50]

    DialogueRNN: An Attentive RNN for Emotion Detection in Conversations , booktitle =

    Majumder, Navonil and Poria, Soujanya and Hazarika, Devamanyu and Mihalcea, Rada and Gelbukh, Alexander and Cambria, Erik , year = 2019, volume =. DialogueRNN: An Attentive RNN for Emotion Detection in Conversations , booktitle =

    2019

  51. [51]

    DialogueTRM: Exploring Multi-Modal Emotional Dynamics in a Conversation , booktitle =

    Mao, Yuzhao and Liu, Guang and Wang, Xiaojie and Gao, Weiguo and Li, Xuan , year = 2021, pages =. DialogueTRM: Exploring Multi-Modal Emotional Dynamics in a Conversation , booktitle =

    2021

  52. [52]

    Ada2I: Enhancing Modality Balance for Multimodal Conversational Emotion Recognition , booktitle =

    Nguyen, Cam-Van Thi and Le, The-Son and Mai, Anh-Tuan and Le, Duc-Trong , year = 2024, month = oct, pages =. Ada2I: Enhancing Modality Balance for Multimodal Conversational Emotion Recognition , booktitle =

    2024

  53. [53]

    Context-Dependent Sentiment Analysis in User-Generated Videos , booktitle =

    Poria, Soujanya and Cambria, Erik and Hazarika, Devamanyu and Majumder, Navonil and Zadeh, Amir and Morency, Louis-Philippe , year = 2017, pages =. Context-Dependent Sentiment Analysis in User-Generated Videos , booktitle =

    2017

  54. [54]

    MELD: A Multimodal Multi-Party Dataset for Emotion Recognition in Conversations , booktitle =

    Poria, Soujanya and Hazarika, Devamanyu and Majumder, Navonil and Naik, Gautam and Cambria, Erik and Mihalcea, Rada , year = 2019, pages =. MELD: A Multimodal Multi-Party Dataset for Emotion Recognition in Conversations , booktitle =

    2019

  55. [55]

    IEEE Access , volume =

    Emotion Recognition in Conversation: Research Challenges, Datasets, and Recent Advances , author =. IEEE Access , volume =

  56. [56]

    IEEE Transactions on Affective Computing , volume =

    Multi-Party Conversation Modeling for Emotion Recognition , author =. IEEE Transactions on Affective Computing , volume =

  57. [57]

    Directed Acyclic Graph Network for Conversational Emotion Recognition , booktitle =

    Shen, Weizhou and Wu, Siyue and Yang, Yunyi and Quan, Xiaojun , year = 2021, pages =. Directed Acyclic Graph Network for Conversational Emotion Recognition , booktitle =

    2021

  58. [58]

    DialogXL: All-in-One XLNet for Multi-Party Conversation Emotion Recognition , booktitle =

    Shen, Weizhou and Chen, Junqing and Quan, Xiaojun and Xie, Zhixian , year = 2021, volume =. DialogXL: All-in-One XLNet for Multi-Party Conversation Emotion Recognition , booktitle =

    2021

  59. [59]

    MultiEMO: An Attention-Based Correlation-Aware Multimodal Fusion Framework for Emotion Recognition in Conversations , booktitle =

    Shi, Tao and Huang, Shao-Lun , year = 2023, pages =. MultiEMO: An Attention-Based Correlation-Aware Multimodal Fusion Framework for Emotion Recognition in Conversations , booktitle =

    2023

  60. [60]

    Efficient Long-Distance Latent Relation-Aware Graph Neural Network for Multi-Modal Emotion Recognition in Conversations , author =

  61. [61]

    2509.24322 , archivePrefix=

    Multimodal Large Language Models Meet Multimodal Emotion Recognition and Reasoning: A Survey , author =. 2509.24322 , archivePrefix=

    arXiv

  62. [62]

    Dynamic Graph Neural ODE Network for Multi-Modal Emotion Recognition in Conversation , booktitle =

    Shou, Yuntao and Meng, Tao and Ai, Wei and Li, Keqin , year = 2025, pages =. Dynamic Graph Neural ODE Network for Multi-Modal Emotion Recognition in Conversation , booktitle =

    2025

  63. [63]

    Well, Now We Know! Unveiling Sarcasm: Initiating and Exploring Multimodal Conversations with Reasoning , booktitle =

    Singh, Gopendra Vikram and Firdaus, Mauajama and Chauhan, Dushyant Singh and Ekbal, Asif and Bhattacharyya, Pushpak , year = 2024, volume =. Well, Now We Know! Unveiling Sarcasm: Initiating and Exploring Multimodal Conversations with Reasoning , booktitle =

    2024

  64. [64]

    Engineering Applications of Artificial Intelligence , pages=

    SUNET: Speaker-Utterance Interaction Graph Neural Network for Emotion Recognition in Conversations , author =. Engineering Applications of Artificial Intelligence , pages=

  65. [65]

    Adaptive Graph Learning for Multimodal Conversational Emotion Detection , booktitle =

    Tu, Geng and Xie, Tian and Liang, Bin and Wang, Hongpeng and Xu, Ruifeng , year = 2024, volume =. Adaptive Graph Learning for Multimodal Conversational Emotion Detection , booktitle =

    2024

  66. [66]

    Contextualized Emotion Recognition in Conversation as Sequence Tagging , booktitle =

    Wang, Yan and Zhang, Jiayu and Ma, Jun and Wang, Shaojun and Xiao, Jing , year = 2020, pages =. Contextualized Emotion Recognition in Conversation as Sequence Tagging , booktitle =

    2020

  67. [67]

    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 , year = 2024, pages =. Emotion Recognition in Conversation via Dynamic Personality , booktitle =

    2024

  68. [68]

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

  69. [69]

    Multimodal Emotion Recognition in Conversations: A Survey of Methods, Trends, Challenges and Prospects , booktitle =

    Wu, ChengYan and Cai, Yiqiang and Liu, Yang and Zhu, Pengxu and Xue, Yun and Gong, Ziwei and Hirschberg, Julia and Ma, Bolei , year = 2025, pages =. Multimodal Emotion Recognition in Conversations: A Survey of Methods, Trends, Challenges and Prospects , booktitle =

    2025

  70. [70]

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

    Yu, Fangxu and Guo, Junjie and Wu, Zhen and Dai, Xinyu , year = 2024, pages =. Emotion-Anchored Contrastive Learning Framework for Emotion Recognition in Conversation , booktitle =

    2024

  71. [71]

    Emotion Detection on TV Show Transcripts with Sequence-Based Convolutional Neural Networks , booktitle =

    Zahiri, Sayyed M and Choi, Jinho D , year = 2018, volume=. Emotion Detection on TV Show Transcripts with Sequence-Based Convolutional Neural Networks , booktitle =

    2018

  72. [72]

    Modeling Both Context- and Speaker-Sensitive Dependence for Emotion Detection in Multi-Speaker Conversations , booktitle =

    Zhang, Dong and Wu, Liangqing and Sun, Changlong and Li, Shoushan and Zhu, Qiaoming and Zhou, Guodong , year = 2019, pages =. Modeling Both Context- and Speaker-Sensitive Dependence for Emotion Detection in Multi-Speaker Conversations , booktitle =

    2019

  73. [73]

    Neural Networks , volume =

    DialogueLLM: Context and Emotion Knowledge-Tuned Large Language Models for Emotion Recognition in Conversations , author =. Neural Networks , volume =

  74. [74]

    CauAIN: Causal Aware Interaction Network for Emotion Recognition in Conversations , booktitle =

    Zhao, Weixiang and Zhao, Yanyan and Lu, Xin , year = 2022, pages =. CauAIN: Causal Aware Interaction Network for Emotion Recognition in Conversations , booktitle =

    2022

  75. [75]

    M3ED: Multi-Modal Multi-Scene Multi-Label Emotional Dialogue Database , booktitle =

    Zhao, Jinming and Zhang, Tenggan and Hu, Jingwen and Liu, Yuchen and Jin, Qin and Wang, Xinchao and Li, Haizhou , year = 2022, pages =. M3ED: Multi-Modal Multi-Scene Multi-Label Emotional Dialogue Database , booktitle =

    2022

  76. [76]

    Knowledge-Enriched Transformer for Emotion Detection in Textual Conversations , booktitle =

    Zhong, Peixiang and Wang, Di and Miao, Chunyan , year = 2019, pages =. Knowledge-Enriched Transformer for Emotion Detection in Textual Conversations , booktitle =

    2019

  77. [77]

    Topic-Driven and Knowledge-Aware Transformer for Dialogue Emotion Detection , booktitle =

    Zhu, Lixing and Pergola, Gabriele and Gui, Lin and Zhou, Deyu and He, Yulan , year = 2021, pages =. Topic-Driven and Knowledge-Aware Transformer for Dialogue Emotion Detection , booktitle =

    2021

  78. [78]

    Artificial Neural Networks and Machine Learning -- ICANN 2024 , pages =

    BiosERC: Integrating Biography Speakers Supported by LLMs for ERC Tasks , author =. Artificial Neural Networks and Machine Learning -- ICANN 2024 , pages =

    2024

  79. [79]

    SKIER: A Symbolic Knowledge Integrated Model for Conversational Emotion Recognition , booktitle =

    Li, Wei and Zhu, Luyao and Mao, Rui and Cambria, Erik , year =. SKIER: A Symbolic Knowledge Integrated Model for Conversational Emotion Recognition , booktitle =

  80. [80]

    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

Showing first 80 references.