Cactus: Towards Psychological Counseling Conversations using Cognitive Behavioral Theory

• DOI: 10.48550/arxiv.2407.03103
• Publication Date: 2024-07-03
• Authors: Suyeon Lee; Sunghwan Kim; Min-Ju Kim; Dongjin Kang; Dongil Yang; Harim Kim; Minseok Kang; Dayi Jung; Min Hee Kim; Seungbeen Lee; Kyoung-Mee Chung; Youngjae Yu; Dongha Lee; Jinyoung Yeo
• Keywords: FOS: Computer and information sciences; Computer Science - Computation and Language; Computation and Language (cs.CL)

Recently, the demand for psychological counseling has significantly increased as more individuals express concerns about their mental health. This surge accelerated efforts to improve accessibility of by using large language models (LLMs) counselors. To ensure client privacy, training open-source LLMs faces a key challenge: absence realistic datasets. address this, we introduce Cactus, multi-turn dialogue dataset that emulates real-life interactions goal-oriented and structured approach Cognitive Behavioral Therapy (CBT). We create diverse designing clients with varied, specific personas, having counselors systematically apply CBT techniques in interactions. assess quality our data, benchmark against established criteria used evaluate real sessions, ensuring alignment expert evaluations. Experimental results demonstrate Camel, model trained outperforms other skills, highlighting its effectiveness potential agent. make model, code publicly available.