Enhancing Emotion-Cause Pair Extraction in Conversation with Contextual Information

This research introduces the Contextualized Pair-Relationship Guided Mixture-of-Experts (CPRG-MoE) model for Emotion–Cause Pair Extraction in Conversation (ECPEC), a task that identifies emotions and their causes in conversation. The model integrates conversational context and dialogue features, addressing previous studies' limitations in utilizing contextual information. A novel evaluation metric, Emotion–Cause Pair Emotion Combined Assessment (PECA), is also proposed, allowing for a comprehensive assessment of emotion-cause pairs and their corresponding emotions. The CPRG-MoE model outperforms baselines on the RECCON and ConvECPE datasets, using both existing and PECA metrics.