PsyDT: Using LLMs to Construct the Digital Twin of Psychological Counselor with Personalized Counseling Style for Psychological Counseling

Currently, large language models (LLMs) have made significant progress in the field of psychological counseling. However, existing mental health LLMs overlook a critical issue where they do not consider the fact that different psychological counselors exhibit different personal styles, including linguistic style and therapy techniques, etc. As a result, these LLMs fail to satisfy the individual needs of clients who seek different counseling styles. To help bridge this gap, we propose PsyDT, a novel framework using LLMs to construct the Digital Twin of Psychological counselor with personalized counseling style. Compared to the timeconsuming and costly approach of collecting a large number of real-world counseling cases to create a specific counselor’s digital twin, our framework offers a faster and more costeffective solution. To construct PsyDT, we utilize dynamic one-shot learning by using GPT-4 to capture counselor’s unique counseling style, mainly focusing on linguistic style and therapy techniques. Subsequently, using existing singleturn long-text dialogues with client’s questions, GPT-4 is guided to synthesize multi-turn dialogues of specific counselor.

Introduction. In recent years, LLMs have made significant advancements, exemplified by ChatGPT (OpenAI, 2022), GPT-4 (OpenAI, 2024), LLaMA (Touvron et al., 2023), Qwen (Bai et al., 2023), ChatGLM (Du et al., 2022), etc. While these LLMs excel in a variety of tasks, they often encounter limitations in specialized fields such as mental health due to a lack of domain-specific expertise. In addition, with the global rise in the prevalence of depression and anxiety (Santomauro et al., 2021), mental health has garnered widespread attention, prompting researchers to explore the application of LLMs in psychological counseling. The value of mental health LLMs lies in their potentiality to provide emotional support and counseling services to individuals. Currently, a series of mental health LLMs have been proposed, including MeChat (Qiu et al., 2024a), PsyChat (Qiu et al., 2024b), SoulChat (Chen et al., 2023), EmoLLM (EmoLLM, 2024), MindChat (Xin Yan, 2023), CPsyCoun (Zhang et al., 2024), etc.

Discussion / Conclusion. In this paper, we propose PsyDT, a novel framework using LLMs to construct the digital twin of psychological counselor with personalized counseling style. The proposed multi-turn dialogues synthesis method of PsyDT framework can quickly and cost-effectively synthesize PsyDTCorpus, a high-quality multi-turn mental health dialogues dataset of psychological counselor with specific counseling style, which closely resemble realworld counseling cases. This indicates the strong potential of PsyDT for application in real-world psychological counseling. Although the experimental results demonstrate the effectiveness of PsyDT, there are still some limitations need to consider. Psychological counseling is complex. Our framework only constructs digital twin of psychological counselor with specific counseling style, which satisfies the individual needs of clients who seek specific counseling style, but can not guarantee to solve their psychological problems and meet counseling needs of all clients.