Evaluating Large Language Models for Psychological Diagnosis and Counseling: A Dual-Task Framework with Cultural Reflection
Large language models (LLMs) have demonstrated high potential in healthcare and mental health related applications, however, their applicability to psychological diagnosis and counseling is not adequately investigated. This paper presents a dual-task assessment model that can be used to systematically evaluate how the mainstream Chinese LLMs can be used in psychology and through two lens psychological diagnosis and psychological counseling. In the diagnosis task, we test four Chinese LLMs, DeepSeek, Doubao, Kimi, and Tongyi Qianwen, on a mental health dataset that consists of five psychological disorders. Accuracy, Precision, Recall, F1-score, AUC, and confusion matrices are the measures of their diagnostic performance. In the case of the counseling task, we develop a prompt-based assessment system and evaluate the generated responses using six dimensions, including Overall, Empathy, Specificity, Medical Advice, Factual Consistency and Toxicity. To lower the expenses of annotating on the large scale with experts, we use an LLM-as-a-Judge approach to rank and evaluate the generated counseling answers. The experimental findings indicate that already existing Chinese LLM has some promising prospects in preliminary psychological diagnosis but fails to differentiate semantically similar disorders resulting in diagnostic confusion. The models themselves tend to produce fluent and empathetic responses in a counseling situation although they display obvious limitations in specificity, safety awareness, and dealing with medical advice. In addition, we also offer a case analysis and cultural reflection, pointing out that good average performance does not always imply the consistent support of culturally diverse adolescent groups. On the whole, our results indicate that the existing LLMs are more appropriate as an assistant, rather than a replacement of a mental health professional, and the research of the future should be aimed at enhancing the safety, cultural sensitivity, and clinical reliability of the system.
References
- Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. (2020). Language models are few-shot learners. In Advances in Neural Information Processing Systems (Vol. 33, pp. 1877–1901).
- Gu, J., Jiang, X., Shi, Z., Tan, H., Zhai, X., Xu, C., Li, W., Shen, Y., Ma, S., Liu, H., Wang, S., Zhang, K., Wang, Y., Gao, W., Ni, L., & Guo, J. (2024). A survey on llm-as-a-judge. arXiv preprint arXiv:2411.15594.
- Guo, Z., Lai, A., Thygesen, J. H., Farrington, J., Keen, T., & Li, K. (2024). Large language models for mental health applications: A systematic review. JMIR Mental Health, 11, e57400. https://doi.org/10.2196/57400
- Jin, Y., Liu, J., Li, P., Wang, B., Yan, Y., Zhang, H., Ni, C., Wang, J., Li, Y., Bu, Y., & Wang, Y. (2025). The applications of large language models in mental health: Scoping review. Journal of Medical Internet Research, 27, e69284. https://doi.org/10.2196/69284
- Nguyen, V. C., Taher, M., Hong, D., Possobom, V. K., Gopalakrishnan, V. T., Raj, E., Li, Z., Soled, H. J., Birnbaum, M. L., Kumar, S., & De Choudhury, M. (2025). Do large language models align with core mental health counseling competencies? In Findings of the Association for Computational Linguistics: NAACL 2025 (pp. 7503–7526).
- Omar, M., Khan, H., Agrawal, A., et al. (2024). Applications of large language models in psychiatry: A systematic review. Frontiers in Psychiatry, 15, 1422807. https://doi.org/10.3389/fpsyt.2024.1422807
- Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. (2022). Training language models to follow instructions with human feedback. In Advances in Neural Information Processing Systems (Vol. 35, pp. 27730–27744).
- Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017). Attention is all you need. In Advances in Neural Information Processing Systems (Vol. 30).
- Xu, Y., Fan, Z., Liu, W., Jin, Y., Ji, W., Bao, P., Wu, J., & Xia, T. (2025). Evaluation of large language models on mental health: From knowledge test to illness diagnosis. Frontiers in Psychiatry, 16, 1646974. https://doi.org/10.3389/fpsyt.2025.1646974
- Zhang, M., et al. (2025). Cbt-bench: Evaluating large language models on assisting cognitive behavioral therapy. In Proceedings of NAACL 2025.