Integrating Emotion and Topic in Empathetic Dialogue from Both Static and Dynamic Perspectives

Jiawang Hao, Fang Kong, Yuanchen Shi

Abstract: Empathetic dialogue has become a pivotal focus in empathy research, emphasizing the application of empathy in communication to establish deeper emotional connections. Previous research has delved deeply into empathetic dialogue from the perspective of user emotions. However, the role of thematic consistency in maintaining meaningful conversations has been overlooked. This paper addresses this research gap by introducing Empathetic Dialogue Generation through Topic-awareness and Emotion Enhancement (EDTE). EDTE leverages commonsense knowledge bases to extract conversational topics and speaker emotions from both static and dynamic perspectives, contributing to both topical relevance and enhanced empathy expression in generated responses. By incorporating thematic consistency, EDTE promotes continuity in conversations, making speakers feel valued and at ease. Experiments on the EmpatheticDialogues dataset demonstrate that EDTE outperforms previous methods in producing responses with stronger topicality and higher empathy.