MindChat-R0: A Large Language Model for Emotionally Supportive Dialogue through Reinforcement Learning

Oct 12, 2025ยท
Dong She
,
Chenxu Zhang
,
Xianrong Yao
,
Yang Gao
,
Zhanpeng Jin
ยท 1 min read
Abstract
MindChat-R0 is a Chinese emotionally supportive dialogue model trained through reinforcement learning with structured emotional reasoning. The work introduces an Empathetic Chain-of-Thought framework and trains the model using reward signals designed to improve empathy, fluency, coherence, and supportive responses for people facing mental health challenges.
Type
Publication
Companion of the 2025 ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp Companion 2025), pp. 1209-1216

Recognition: 10th Mental Health Workshop Best Paper Honorable Mention

UbiComp Mental Health Emotionally Supportive Dialogue Large Language Models Reinforcement Learning Human-Centered AI