EmoStage: A Framework for Accurate Empathetic Response Generation via Perspective-Taking and Phase Recognition

The rising demand for mental health care has fueled interest in AI-driven counseling systems. While large language models (LLMs) offer significant potential, current approaches face challenges, including limited understanding of clients' psychological states and counseling stages, reliance on high-quality training data, and privacy concerns associated with commercial deployment. To address these issues, we propose EmoStage, a framework that enhances empathetic response generation by leveraging the inference capabilities of open-source LLMs without additional training data. Our framework introduces perspective-taking to infer clients' psychological states and support needs, enabling the generation of emotionally resonant responses. In addition, phase recognition is incorporated to ensure alignment with the counseling process and to prevent contextually inappropriate or inopportune responses. Experiments conducted in both Japanese and Chinese counseling settings demonstrate that EmoStage improves the quality of responses generated by base models and performs competitively with data-driven methods.

Can Large Language Models be Used to Provide Psychological Counselling? An Analysis of GPT-4-Generated Responses Using Role-play Dialogues

Mental health care poses an increasingly serious challenge to modern societies. In this context, there has been a surge in research that utilizes information technologies to address mental health problems, including those aiming to develop counseling dialogue systems. However, there is a need for more evaluations of the performance of counseling dialogue systems that use large language models. For this study, we collected counseling dialogue data via role-playing scenarios involving expert counselors, and the utterances were annotated with the intentions of the counselors. To determine the feasibility of a dialogue system in real-world counseling scenarios, third-party counselors evaluated the appropriateness of responses from human counselors and those generated by GPT-4 in identical contexts in role-play dialogue data. Analysis of the evaluation results showed that the responses generated by GPT-4 were competitive with those of human counselors.