본문 바로가기
  • Home

Integrating Emotion, Situation, and Empathy Strategies in Retrieval-Augmented Response Generation

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2026, 31(7), pp.55~64
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : May 13, 2026
  • Accepted : July 1, 2026
  • Published : July 31, 2026

So-Jeong Park 1 Beak-Cheol Jang 1

1연세대학교

Accredited

ABSTRACT

Empathetic Response Generation aims to produce appropriate responses by understanding a speaker’s emotions and situational context, and it requires a balanced integration of affective and cognitive empathy. Prior studies have selectively addressed emotion recognition, situation/cause modeling, and empathy strategies. Recent retrieval-augmented approaches retrieve examples mainly by semantic similarity and treat strategy selection as a separately fine-tuned classifier. In this paper, we propose a RAG-based framework that extracts emotion and situation from user utterances and performs two-stage retrieval, consisting of situation-embedding-based initial retrieval and emotion-based filtering refinement; the empathy-strategy labels of the retrieved examples are used directly as few-shot context, without a separate strategy-prediction model.

Citation status

* References for papers published after 2025 are currently being built.