Multimodal system improves counselor responses with reasoning consistency
MOCC-R1: Reinforcing Reasoning-Response Consistency for Multimodal Counselor Response Generation
Artificial Intelligence
Summary
Counselor chatbots try to respond helpfully using both words and other information like images or emotions, but existing tools struggle to keep their replies consistent with the reasoning behind them. The authors created a large dataset of real counseling conversations with verified counselors to help improve this. They also built a system that first plans a response based on understanding the client and counseling principles, then uses reinforcement learning to make sure the final reply matches the plan. This approach makes counselor responses more reliable and relevant.
What this means in practice
- •For mental health app developers: Generate more consistent and context-aware chatbot counselor responses based on real counseling data to improve automated mental health support.$Commercial implications: Enables development of commercial mental health chatbots with verified counselor dialogue history and improved reasoning alignment.
- •For customer support platform teams: Enhance AI chat systems to produce responses that better align with planned reasoning for more reliable support interactions.
Authors
Wenjie Zheng, Qiming Xie, Jianfei Yu, Rui Xia
Abstract
Multimodal counselor response generation (MCRG) aims to generate an appropriate counselor response from multimodal dialogue histories. Progress is limited by two gaps: first, existing datasets rarely capture sustained, human-recorded counseling interactions conducted by qualified counselors; Second, existing methods do not explicitly optimize consistency between counseling reasoning and the generated response, potentially undermining the reliability of MCRG systems. Thus, we introduce MOCC, a multimodal counseling conversation corpus containing over 200 hours of interactions involving 154 credential-verified counselors. Based on MOCC, we propose MOCC-R1, a two-stage framework for optimizing reasoning-response consistency. Cold-start supervised fine-tuning trains the model to generate a structured trajectory consisting of client-state understanding, a response intent that links a counseling principle to a planned action, and the final response. Reinforcement learning (RL) then rewards grounded plan coherence and plan execution, encouraging the inferred state and plan to be supported by the dialogue context and the response to realize that plan. Experiments demonstrate the effectiveness of the proposed MOCC-R1.