Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Following the TRACE: A Structured Path to Empathetic Response Generation with Multi-Agent Models

Ziqi Liu, Ziyang Zhou, Yilin Li, Haiyang Zhang, Yangbin Chen

TL;DR

Empathetic response generation faces a trade-off between analytical depth and linguistic fluency. TRACE introduces a four-agent pipeline—Affective State Identifier, Causal Analysis Engine, Strategic Response Planner, and Empathetic Response Synthesizer—that decomposes empathy into perception, reasoning, planning, and synthesis, aided by retrieval-augmented components. It achieves strong automatic and human-like performance, with notable gains in diversity and emotion accuracy (I-ACC) and demonstrable interpretability from its modular design. This structured approach yields more targeted, coherent, and expressive empathetic responses, offering a practical path to bridging specialized analytical models and large language models in real-world conversational agents.

Abstract

Empathetic response generation is a crucial task for creating more human-like and supportive conversational agents. However, existing methods face a core trade-off between the analytical depth of specialized models and the generative fluency of Large Language Models (LLMs). To address this, we propose TRACE, Task-decomposed Reasoning for Affective Communication and Empathy, a novel framework that models empathy as a structured cognitive process by decomposing the task into a pipeline for analysis and synthesis. By building a comprehensive understanding before generation, TRACE unites deep analysis with expressive generation. Experimental results show that our framework significantly outperforms strong baselines in both automatic and LLM-based evaluations, confirming that our structured decomposition is a promising paradigm for creating more capable and interpretable empathetic agents. Our code is available at https://anonymous.4open.science/r/TRACE-18EF/README.md.

Following the TRACE: A Structured Path to Empathetic Response Generation with Multi-Agent Models

TL;DR

Empathetic response generation faces a trade-off between analytical depth and linguistic fluency. TRACE introduces a four-agent pipeline—Affective State Identifier, Causal Analysis Engine, Strategic Response Planner, and Empathetic Response Synthesizer—that decomposes empathy into perception, reasoning, planning, and synthesis, aided by retrieval-augmented components. It achieves strong automatic and human-like performance, with notable gains in diversity and emotion accuracy (I-ACC) and demonstrable interpretability from its modular design. This structured approach yields more targeted, coherent, and expressive empathetic responses, offering a practical path to bridging specialized analytical models and large language models in real-world conversational agents.

Abstract

Empathetic response generation is a crucial task for creating more human-like and supportive conversational agents. However, existing methods face a core trade-off between the analytical depth of specialized models and the generative fluency of Large Language Models (LLMs). To address this, we propose TRACE, Task-decomposed Reasoning for Affective Communication and Empathy, a novel framework that models empathy as a structured cognitive process by decomposing the task into a pipeline for analysis and synthesis. By building a comprehensive understanding before generation, TRACE unites deep analysis with expressive generation. Experimental results show that our framework significantly outperforms strong baselines in both automatic and LLM-based evaluations, confirming that our structured decomposition is a promising paradigm for creating more capable and interpretable empathetic agents. Our code is available at https://anonymous.4open.science/r/TRACE-18EF/README.md.

Paper Structure

This paper contains 18 sections, 7 equations, 1 figure, 3 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Modeling Empathetic Understanding
  4. Generative Fluency and The Synthesis Challenge
  5. METHODOLOGY
  6. Overall Pipeline
  7. Affective State Identifier
  8. Causal Analysis Engine
  9. Strategic Response Planner
  10. Empathetic Response Synthesizer
  11. EXPERIMENTS
  12. Experimental Setup
  13. Main Result
  14. Automatic Evaluation Results
  15. Human-like Evaluation Results
  16. ...and 3 more sections

Figures (1)

  • Figure 1: The overall architecture of our TRACE framework. The system processes a dialogue history through a four-agent pipeline, where each agent enriches the context with a specific layer of analysis before the final response is synthesized. Agents 3 and 4 leverage a RAG system to consult a knowledge base of examples.