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

TL;DR

PsyDT addresses the mismatch between generic mental health LLMs and the need for counselor-specific styles by constructing a Digital Twin of a psychologist with a personalized counseling style. It combines dynamic one-shot learning using GPT-4 to capture linguistic style and therapy techniques, client personality simulation via Big Five traits, and multi-turn dialogue synthesis to generate a high-quality synthetic corpus (PsyDTCorpus). The model PsyDTLLM is then fine-tuned on this synthetic data using multi-turn instruction fine-tuning with loss $\ell_{\theta}$, achieving superior performance over baselines in both automatic and human evaluations and demonstrating strong potential for real-world personalized psychological counseling. The approach offers a faster, cost-effective alternative to collecting large volumes of real-world counseling data and facilitates practical deployment of tailored counseling LLMs while acknowledging ethical and safety considerations. Overall, PsyDT advances personalized mental health AI by enabling digital twins that reflect individual counselors' styles and techniques with demonstrated effectiveness in synthetic multi-turn dialogues and evaluative results.

Abstract

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 time-consuming 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 cost-effective 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 single-turn long-text dialogues with client's questions, GPT-4 is guided to synthesize multi-turn dialogues of specific counselor. Finally, we fine-tune the LLMs on the synthetic dataset, PsyDTCorpus, to achieve the digital twin of psychological counselor with personalized counseling style. Experimental results indicate that our proposed PsyDT framework can synthesize multi-turn dialogues that closely resemble real-world counseling cases and demonstrate better performance compared to other baselines, thereby show that our framework can effectively construct the digital twin of psychological counselor with a specific counseling style.

Figures (31)

• Figure 1: Difference between our proposed PsyDT framework and others. (a) Previous methods mixed multi-turn dialogues with multiple counseling styles to fine-tune LLM. (b) Our proposed PsyDT framework uses LLMs to construct the digital twin of psychological counselor with a specific counseling style.
• Figure 2: Illustration of multi-turn dialogues synthesis method of PsyDT framework and PsyDTLLM model.
• Figure 3: Distribution of counseling topics.
• Figure 4: Similarity results for the proposed multi-turn dialogue synthesis method and other baseline methods.
• Figure 5: Results of manual evaluation for PsyDTCorpus and baseline datasets on 4 professional dimensions.