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

2026-07-29

AI Summary

This work addresses the challenge in conversation emotion recognition (ERC) that global contextual heterogeneity and irrelevant information often hinder effective modeling of the overall emotional atmosphere. To this end, the authors propose AtmosERC, a novel framework that explicitly constructs an utterance-speaker graph structure and employs a relation-aware graph neural network to extract dialogue-level and speaker-conditioned emotional priors. These priors can either serve as guiding signals for lightweight sequential models or be injected as prompts into large language models, enhancing performance without modifying their backbone architectures. Experimental results demonstrate that AtmosERC significantly improves emotion recognition across four ERC benchmarks, offering both plug-and-play flexibility and robustness to local emotional fluctuations.

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.

Problem

• Emotion Recognition in Conversation
• affective atmosphere
• context modeling
• dialogue-level emotion
• heterogeneous context

Innovation

• affective atmosphere
• graph-based modeling
• emotion recognition in conversation
• relation-aware graph extractor
• LLM prompting