PsyCLIENT: Client Simulation via Conversational Trajectory Modeling for Trainee Practice and Model Evaluation in Mental Health Counseling

TL;DR

PsyCLIENT addresses the need for diverse, realistic, and multilingual client simulations for training and evaluation in mental health counseling by introducing conversational trajectory modeling. It formalizes a trajectory-conditioned generation framework, couples a Chinese client-profile dataset (PsyCLIENT-CP) with trajectory extraction, and demonstrates that conditioning LLM-based clients on explicit behavior trajectories yields dialogues with high authenticity and training value. Empirical evaluations with licensed counselors show PsyCLIENT outperforms strong baselines in authenticity and effectiveness, while expert discrimination tasks indicate dialogues are nearly indistinguishable from human interactions (approximately 95% confusion rate). The work provides an open-source foundation for scalable, culturally grounded client simulation and offers a principled path for evaluating LLM-based counselors, though it is limited to single-session interactions and relies on Chinese cultural contexts that may require adaptation for broader international deployment.

Abstract

LLM-based client simulation has emerged as a promising tool for training novice counselors and evaluating automated counseling systems. However, existing client simulation approaches face three key challenges: (1) limited diversity and realism in client profiles, (2) the lack of a principled framework for modeling realistic client behaviors, and (3) a scarcity in Chinese-language settings. To address these limitations, we propose PsyCLIENT, a novel simulation framework grounded in conversational trajectory modeling. By conditioning LLM generation on predefined real-world trajectories that incorporate explicit behavior labels and content constraints, our approach ensures diverse and realistic interactions. We further introduce PsyCLIENT-CP, the first open-source Chinese client profile dataset, covering 60 distinct counseling topics. Comprehensive evaluations involving licensed professional counselors demonstrate that PsyCLIENT significantly outperforms baselines in terms of authenticity and training effectiveness. Notably, the simulated clients are nearly indistinguishable from human clients, achieving an about 95\% expert confusion rate in discrimination tasks. These findings indicate that conversational trajectory modeling effectively bridges the gap between theoretical client profiles and dynamic, realistic simulations, offering a robust solution for mental health education and research. Code and data will be released to facilitate future research in mental health counseling.

Figures (35)

• Figure 1: Illustration of our automatic client simulation via conversational trajectory modeling. Step 1 is dataset construction. Step 2 is conversational trajectory extraction. The utterances spoken are abstracted to behavior labels and anonymous content. There are 12 types of client behavior labels, including "pi: Providing Information", "co: Confirming", "ec: Expressing Confusion", "de: Defending", etc. (see Figure \ref{['Fig-client-behaviors']} in Appendix \ref{['sec:client-behaviors']}). For a detailed illustration, see Figure \ref{['Fig-method']} in Appendix \ref{['app:simulation-framework']}.
• Figure 2: The topic diversity of PsyCLIENT-CP. The values in parentheses indicate the number of times this topic appears in the PsyCLIENT-CP dataset.
• Figure 3: Discrimination accuracy of human expert evaluation.
• Figure 4: The detailed illustration of our automatic client simulation framework via conversational trajectory modeling.
• Figure 5: An example of an AI client. For the English version, see Figure \ref{['Fig-ai-client-en']}.