CARE: Cognitive-reasoning Augmented Reinforcement for Emotional Support Conversation
Jie Zhu, Yuanchen Zhou, Shuo Jiang, Junhui Li, Lifan Guo, Feng Chen, Chi Zhang, Fang Kong
2025-10-08
Summary
This paper introduces a new method, called CARE, for improving how well computer programs can provide emotional support in conversations, focusing on making their responses more logical and helpful.
What's the problem?
Current approaches to building emotional support chatbots often rely on creating lots of artificial conversation data, but they don't really focus on teaching the chatbot *how* to think through the conversation and provide truly reasoned support. They miss the core cognitive skills needed for effective emotional support, like understanding what someone is saying and responding in a way that makes sense.
What's the solution?
The researchers developed CARE, which works by helping the chatbot learn to reason better using the existing emotional support conversation data it already has. It doesn't need huge amounts of new, fake data. CARE guides the chatbot to generate responses that are logically connected to what the user said and are genuinely supportive. Then, they used a technique called reinforcement learning to further refine this reasoning process, essentially rewarding the chatbot for giving good responses.
Why it matters?
This work is important because it moves beyond simply creating more data and instead focuses on improving the *quality* of the chatbot's thinking. By making chatbots more logically sound and supportive, we can create systems that are more empathetic, helpful, and feel more human-like when providing emotional support, which could be really valuable for people struggling with stress or difficult emotions.
Abstract
Emotional Support Conversation (ESC) plays a vital role in alleviating psychological stress and providing emotional value through dialogue. While recent studies have largely focused on data augmentation and synthetic corpus construction, they often overlook the deeper cognitive reasoning processes that underpin effective emotional support. To address this gap, we propose CARE, a novel framework that strengthens reasoning in ESC without relying on large-scale synthetic data. CARE leverages the original ESC training set to guide models in generating logically coherent and supportive responses, thereby explicitly enhancing cognitive reasoning. Building on this foundation, we further employ reinforcement learning to refine and reinforce the reasoning process. Experimental results demonstrate that CARE significantly improves both the logical soundness and supportive quality of responses, advancing the development of empathetic, cognitively robust, and human-like emotional support systems.