A Strategic Questioning Framework for AI Counseling using Large Language Models and Deep Reinforcement Learning.
The growing demand for mental health services highlighted the need for AI-powered chatbots. However, existing counselor chatbots face significant shortcomings, including reliance on predetermined scripts, client-led conversation models, and a lack of implementation details for advanced features, restricting their ability to handle complex emotional contexts, guide scientifically validated counseling processes, and scale effectively.
We propose Deep SAGE, a Strategic AI Guidance Engine (SAGE). Deep SAGE integrates essential counselor characteristics with a scientifically validated counseling process, specifically cognitive behavioral therapy (CBT). Leveraging advanced natural language processing (NLP), large language models (LLMs), and deep reinforcement learning (DRL), fostering deeper self-disclosure and openness through strategic questioning and context-aware responses. Our work bridges the gap between theoretical counseling frameworks and practical, scalable implementations, offering a cost-effective and accessible mental health resource.
Publications
In Preparation.
Authors: Qi Zhang, Heajun An, Minqian Liu, Xinyi Zhang, Sang Won Lee, Lifu Hwang, Pamela Wisniewski and Jin-Hee Cho
Presentations
ACM CAPWIC 2025 – Student Research Short Talk
Presented “Deep SAGE: A Strategic Questioning Framework for AI Counseling using Large Language Models and Deep Reinforcement Learning” at ACM CAPWIC 2025, March 2025.