PsychēChat: An Empathic Framework Focused on Emotion Shift Tracking and Safety Risk Analysis in Psychological Counseling

TL;DR

PsychēChat addresses the need for explicit emotion-shift tracking and proactive safety risk analysis in AI-assisted psychological counseling. It introduces an interactive role-playing framework with an Emotion Management Module and a Risk Control Module, plus two inference paradigms (Agent Mode and LLM Mode) to balance interpretability and efficiency within an Emotion-Focused Therapy framework. Across SAGE, ESC-Eval, and PsychēEval, PsychēChat demonstrates improved emotional understanding, reduced risk, and strong human-evaluated empathy and safety, outperforming existing psychological LLMs and approaching closed-source upper bounds. The work provides a practical path toward safer, more context-aware AI counselors, with a publicly described dataset (PsychēDialog) and a scalable synthesis pipeline, while acknowledging limitations like generation cost and cultural scope.

Abstract

Large language models (LLMs) have demonstrated notable advancements in psychological counseling. However, existing models generally do not explicitly model seekers' emotion shifts across counseling sessions, a core focus in classical psychological schools. Moreover, how to align counselor models' responses with these emotion shifts while proactively mitigating safety risks remains underexplored. To bridge these gaps, we propose PsychēChat, which explicitly integrates emotion shift tracking and safety risk analysis for psychological counseling. Specifically, we employ interactive role-playing to synthesize counselor--seeker dialogues, incorporating two modules: Emotion Management Module, to capture seekers' current emotions and emotion shifts; and Risk Control Module, to anticipate seekers' subsequent reactions and identify potential risks. Furthermore, we introduce two modeling paradigms. The Agent Mode structures emotion management, risk control, and counselor responses into a collaborative multi-agent pipeline. The LLM Mode integrates these stages into a unified chain-of-thought for end-to-end inference, balancing efficiency and performance. Extensive experiments, including interactive scoring, dialogue-level evaluation, and human assessment, demonstrate that PsychēChat outperforms existing methods for emotional insight and safety control.

Figures (15)

• Figure 1: Two Core Focuses of PsychēChat. (a) Tracking emotion shifts helps generate more context-aware and empathetic responses. (b) Different counselor responses can lead to different levels of safety risk.
• Figure 2: The overview of PsychēChat. The upper part shows an interactive dialogue synthesis framework with two modules for emotion management and risk control. The lower part presents two inference paradigms, Agent Mode and LLM Mode, designed to balance efficiency and performance.
• Figure A1: Topic distribution of PsychēDialog.
• Figure :
• Figure A2: The prompt for generating the role card.