Empathetic Dialogue Generation via Sensitive Emotion Recognition and Sensible Knowledge Selection

• DOI: 10.18653/v1/2022.findings-emnlp.340
• Publication Date: 2023-08-04T20:21:02Z
• Authors: Lanrui Wang; Jiangnan Li; Zheng Lin; Fandong Meng; Chenxu Yang; Weiping Wang; Jie Zhou
• Keywords: Utterance

Empathy, which is widely used in psychological counseling, a key trait of everyday human conversations. Equipped with commonsense knowledge, current approaches to empathetic response generation focus on capturing implicit emotion within dialogue context, where the emotions are treated as static variable throughout However, change dynamically between utterances, makes previous works difficult perceive flow and predict correct target response, leading inappropriate response. Furthermore, simply importing knowledge without harmonization may trigger conflicts emotion, confuse model choose information guide process. To address above problems, we propose Serial Encoding Emotion-Knowledge interaction (SEEK) method for generation. We use fine-grained encoding strategy more sensitive dynamics (emotion flow) conversations emotion-intent characteristic Besides, design novel framework solve generate sensible Extensive experiments utterance-level annotated EMPATHETICDIALOGUES demonstrate that SEEK outperforms strong baseline both automatic manual evaluations.