Elamur Research⌕
☰
home / papers / 2025 / PsyDT: Using LLMs to Construct the Digital Twin of Psychological Counselor with Personalized Counseling Style for Psychological Counseling

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

Paper · 2025 · Haojie Xie, Yirong Chen, Xiaofen Xing et al.

Summary

Mental health LLMs are usually trained on dialogue data that mixes many counselors. As a result, they cannot reproduce how one particular therapist works, and the blended styles can make their answers inconsistent. PsyDT builds a "digital twin" of a single counselor without collecting large numbers of that person's real sessions. The authors recorded 12 text-based sessions between a certified counselor and volunteer clients, one session per topic. GPT-4 extracted the counselor's linguistic style and therapy approach from these cases, and a topic-matched case then served as a one-shot example during generation. Separately, GPT-4 inferred Big Five personality profiles for the clients in 5,000 long single-turn Q&A pairs from SoulChatCorpus. It then expanded each pair into a multi-turn conversation in the target counselor's style. The resulting Chinese dataset, PsyDTCorpus, was used for multi-turn instruction fine-tuning (MIFT) to produce PsyDTLLM. This model outperforms general-purpose and mental-health LLMs on automatic overlap metrics and on expert-rated counseling quality. The approach offers a cheap route to style-specific counseling assistants, although it currently relies on a single counselor.

Key points

  • Seed data: 5,000 high-quality single-turn long-text dialogues were drawn from SoulChatCorpus. Each has a topic, a question title, question details and a long counselor reply. All data is in Chinese.
  • Real-world cases: 12 sessions of 50 minutes each covered 12 topics and cost about ¥12,000. All 12 sessions used Rational Emotive Behavior Therapy (REBT), which was then used to retrieve techniques from a therapy knowledge base.
  • Results: PsyDTLLM achieved the best scores on every automatic metric, with R-1 36.03, R-2 10.08, R-L 28.86, B-4 9.99 and BERTScore 96.79. The next-best R-1 was ChatGPT at 31.72.
  • Professional ratings: PsyDTLLM also led the expert evaluation, scoring 1.90 on emotional empathy, 2.13 on cognitive empathy, 2.19 on conversation strategy and 2.26 on state/attitude. Every model scored 1.00 on safety.
  • Baselines and availability: Baselines included MeChat, PsyChat, SoulChat, MindChat, EmoLLM, CPsyCounX, GPT-4 and several open 7–9B chat models. Code and data are available at github.com/scutcyr/SoulChat2.0.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓South China University of TechnologyGuangzhou · China22 works · 1,554 citations

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksPsychology · 1,552 worksExperimental and Cognitive Psychology · 870 worksApplied Psychology · 21 works

Cite

Cite as Haojie Xie, Yirong Chen, Xiaofen Xing, Jingkai Lin and Xiangmin Xu. 2025. PsyDT: Using LLMs to Construct the Digital Twin of Psychological Counselor with Personalized Counseling Style for Psychological Counseling. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.acl-long.55

@inproceedings{xie2025psydt,
  title = {{PsyDT: Using LLMs to Construct the Digital Twin of Psychological Counselor with Personalized Counseling Style for Psychological Counseling}},
  author = {Haojie Xie, Yirong Chen, Xiaofen Xing, Jingkai Lin and Xiangmin Xu},
  booktitle = {Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2025},
  doi = {10.18653/v1/2025.acl-long.55},
  url = {https://doi.org/10.18653/v1/2025.acl-long.55}
}

References

38 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Data Privacy: We implemented a rigorous data cleaning protocol to synthesize our dataset and ensure privacy protection (Hovy and Spruit, 2016). These measures included rule-based cleaning, manual rewriting, and human proofreading to guarantee the absence of sensitive or private content. For instance… find
  2. Potential Risks of the Model: During the human evaluation phase, we conducted a focused safety assessment of the model’s outputs. Given the absence of human feedback during the model fine-tuning phase, some responses may inevitably pose potential harm to users. If there is no noticeable improvement… find
  3. Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, et al. 2023. Qwen technical report. Preprint, arXiv:2309.16609.
  4. Aaron T Beck. 1979. Cognitive therapy of depression. Guilford press. find
  5. Zheng Cai, Maosong Cao, Haojiong Chen, Kai Chen, Keyu Chen, et al. 2024. Internlm2 technical report. Preprint, arXiv:2403.17297.
  6. Raymond B Cattell. 1946. The description andmeasure-ment of personality. Yonkers-on-Hudson, NY: World. find
  7. Yirong Chen, Xiaofen Xing, Jingkai Lin, Huimin Zheng, Zhenyu Wang, Qi Liu, and Xiangmin Xu. 2023. SoulChat: Improving LLMs’ empathy, listening, and comfort abilities through fine-tuning with multi-turn empathy conversations. In Findings of the Association for Computational Linguistics: EMNLP 2023, p… find
  8. Cheng-Han Chiang and Hung-yi Lee. 2023. Can large language models be an alternative to human evaluations? In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Toronto, Canada. Association for Computational Linguistics. find
  9. Paul T Costa and Robert R McCrae. 1999. A five-factor theory of personality. The five-factor model of personality: Theoretical perspectives, 2:51–87. find
  10. Windy Dryden. 2005. Rational emotive behavior therapy. Encyclopedia of cognitive behavior therapy, pages 321–324. find
  11. Zhengxiao Du, Yujie Qian, Xiao Liu, Ming Ding, Jiezhong Qiu, Zhilin Yang, and Jie Tang. 2022. GLM: General language model pretraining with autoregressive blank infilling. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 320–335…
  12. Joseph L Fleiss. 1971. Measuring nominal scale agreement among many raters. Psychological Bulletin, 76(5):378–382. find
  13. Team GLM, :, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, et al. 2024. Chatglm: A family of large language models from glm-130b to glm-4 all tools. Preprint, arXiv:2406.12793.
  14. Allen L Hammer. 1996. MBTI applications: A decade of research on the Myers-Briggs Type Indicator. Consulting Psychologists Press. find
  15. Lucia Hernawati, Bagus Wismanto, and Rachmad Djati Winarno. 2019. The differences of counseling satisfaction between counselees with counseling style matches with their personalities and those who are not. Open Universiteit www. ou. nl Cyberjaya University College of Medical Science De La Salle Univ… find
  16. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311–318, Philadelphia, Pennsylvania, USA. Association for Computational…
  17. Nancy Pinchot, Anthony C Riccio, and Herman J Peters. 1975. Elementary school students’ and their parents’ preferences for counselors. Counselor Education and Supervision, 15(1):28–33. find
  18. Stephanie D Preston and Frans BM De Waal. 2002. Empathy: Its ultimate and proximate bases. Behavioral and brain sciences, 25(1):1–20. find
  19. Huachuan Qiu, Hongliang He, Shuai Zhang, Anqi Li, and Zhenzhong Lan. 2024a. SMILE: Single-turn to multi-turn inclusive language expansion via ChatGPT for mental health support. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 615–636, Miami, Florida, USA. Association f…
  20. Huachuan Qiu, Anqi Li, Lizhi Ma, and Zhenzhong Lan. 2024b. Psychat: A client-centric dialogue system for mental health support. In 2024 27th International Conference on Computer Supported Cooperative Work in Design (CSCWD), pages 2979–2984. find
  21. Dirk Hovy and Shannon L. Spruit. 2016. The social impact of natural language processing. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 591–598, Berlin, Germany. Association for Computational Linguistics.
  22. Damian F Santomauro, Ana M Mantilla Herrera, Jamileh Shadid, Peng Zheng, Charlie Ashbaugh, et al. 2021. Global prevalence and burden of depressive and anxiety disorders in 204 countries and territories in 2020 due to the covid-19 pandemic. The Lancet, 398(10312):1700–1712. find
  23. Chin-Yew Lin. 2004. ROUGE: A package for automatic evaluation of summaries. In Text Summarization Branches Out, pages 74–81, Barcelona, Spain. Association for Computational Linguistics.
  24. Zhuang She, Tomáš Řiháček, Jun Xu, Wenxian Yang, Dan Xu, Ningning Zhou, Weidong Ji, and Juzhe Xi. 2023. Psychometric evaluation of the cooper– norcross inventory of preferences–therapist version. Assessment, 30(5):1651–1661. find
  25. Marsha Linehan. 2014. DBT? Skills training manual. Guilford Publications. find
  26. Gerald Matthews, Ian J Deary, and Martha C White-man. 2003. Personality traits. Cambridge University Press. find
  27. Augustine Meier. 2010. Counselling and therapy techniques: Theory & practice. find
  28. OpenAI. 2022. Introducing chatgpt. find
  29. OpenAI. 2024. Gpt-4 technical report. arXiv:2303.08774.
  30. Preprint, Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, et al. 2022. Training language models to follow instructions with human feedback. In Advances in Neural Information Processing Systems, volume 35, pages 27730–27744. Curran Associates, Inc.
  31. Jacqueline M Swank and Glenn W Lambie. 2012. The assessment of cacrep core curricular areas and student learning outcomes using the counseling competencies scale. Counseling Outcome Research and Evaluation, 3(2):116–127. find
  32. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023. Llama: Open and efficient foundation language models. Preprint…
  33. Dong Xue* Xin Yan. 2023. Mindchat: Psychological large language model. https://github.com/ X-D-Lab/MindChat. ↗
  34. Aiyuan Yang, Bin Xiao, Bingning Wang, Borong Zhang, Ce Bian, et al. 2023. Baichuan 2: Open large-scale language models. Preprint, arXiv:2309.10305.
  35. An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, et al. 2024. Qwen2 technical report. Preprint, arXiv:2407.10671. find
  36. Alex Young, Bei Chen, Chao Li, Chengen Huang, et al. 2024. Yi: Open foundation models by 01.ai. Preprint, arXiv:2403.04652.
  37. Chenhao Zhang, Renhao Li, Minghuan Tan, Min Yang, Jingwei Zhu, Di Yang, Jiahao Zhao, Guancheng Ye, Chengming Li, and Xiping Hu. 2024. CPsyCoun: A report-based multi-turn dialogue reconstruction and evaluation framework for Chinese psychological counseling. In Findings of the Association for Computat… find
  38. Tianyi Zhang*, Varsha Kishore*, Felix Wu*, Kilian Q. Weinberger, and Yoav Artzi. 2020. bertbertscore: Evaluating text generation with bert. In International Conference on Learning Representations. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xie-2025-psydt-llms-construct-digital-twin/