Dual-role interaction mechanism-based large language model for psychological health

doi:10.11772/j.issn.1001-9081.2025060783

Journal of Computer Applications ›› 2026, Vol. 46 ›› Issue (7): 2054-2063.DOI: 10.11772/j.issn.1001-9081.2025060783

• Artificial intelligence • Previous Articles

Dual-role interaction mechanism-based large language model for psychological health

Yadian CHENG¹, Yingying LI²(), Ping ZHANG¹, Fangbing QIU³, Xiaonan CHAI³, Yuqiao SHU³

^1.School of Cyberspace Security，Information Engineering University，Zhengzhou Henan 450001，China
^2.National Research Center of Parallel Computer Engineering Technology，Jiangnan Institute of Computing Technology，Wuxi Jiangsu 214000，China
^3.National Supercomputing Center in Zhengzhou （Zhengzhou University），Zhengzhou Henan 450001，China

Received:2025-07-16 Revised:2025-10-23 Accepted:2025-10-29 Online:2025-11-27 Published:2026-07-10
Contact: Yingying LI
About author:CHENG Yadian， born in 2002， M. S. candidate. Her research interests include large language models， advanced compilation.
ZHANG Ping， born in 1969， Ph. D.， professor. Her research interests include advanced compilation， information system security.
QIU Fangbing， born in 1995， M. S. Her research interests include large language models， deep learning.
CHAI Xiaonan， born in 1996， M. S. Her research interests include deep learning compilation， algorithm parallelization.
SHU Yuqiao， born in 1998， M. S. Her research interests include large language models， operator optimization.
Supported by:
Advanced Computing and Intelligent Engineering Laboratory Project(1017406)

基于双角色交互机制的心理健康大语言模型

程雅典¹, 李颖颖²(), 张平¹, 邱芳冰³, 柴晓楠³, 舒玉巧³

^1.信息工程大学网络空间安全学院，郑州 450001
^2.国家并行计算机工程技术研究中心（江南计算技术研究所），江苏无锡 214000
^3.国家超级计算郑州中心（郑州大学），郑州 450001

通讯作者: 李颖颖
作者简介:程雅典（2002—），女，河南新乡人，硕士研究生，CCF会员，主要研究方向：大语言模型、先进编译
张平（1969—），女，吉林通化人，教授，博士，主要研究方向：先进编译、信息系统安全
邱芳冰（1995—），女，河南南阳人，硕士，主要研究方向：大语言模型、深度学习
柴晓楠（1996—），女，河南新乡人，硕士，主要研究方向：深度学习编译、算法并行
舒玉巧（1998—），女，河南信阳人，硕士，主要研究方向：大语言模型、算子优化。
基金资助:
先进计算与智能工程实验室项目(1017406)

Abstract

Abstract:

Existing Artificial Intelligence （AI） psychological counseling systems are generally limited to the counselor role， making it difficult to address users' dual needs for both professional advice and emotional support dynamically. To overcome this limitation， a psychological health Large Language Model （LLM） based on dual-role interaction mechanism， named STAR （Supportive Therapeutic Adaptive Responder）， was proposed. First， a dynamic role switching mechanism based on system prompts was designed to enable stylistic shifts between the psychological counselor and friend roles， thereby addressing users' dual needs for both professional advice and emotional support in psychological counseling scenarios. Then， a training process incorporating data quality control was implemented， where model feedback was continuously collected to dynamically optimize both data generation strategies and system prompt design， thereby forming a closed loop between data generation and model training and enhancing dataset quality. Experimental results on the Chinese psychology benchmark testing framework CPsyCoun （Chinese Psychological Counseling） show that the STAR model achieves improvements of 9.80% in comprehensive score of comprehensiveness， professionalism， and authenticity over the base model Qwen2.5-7B-Instruct， and 26.81% over the single-role psychological LLM EmoLLM （Emotional Large Language Model）. STAR model also outperforms models fine-tuned on open-source datasets such as SoulChat， as well as general-purpose LLMs like GPT-4o-mini. The STAR model achieves dual-role adaptive responding effectively， significantly enhancing both effect and experience of psychological counseling.

Key words: Large Language Model (LLM), psychological counseling, dual-role interaction, emotional support, instruction fine-tuning

摘要：

针对现有的人工智能（AI）心理咨询系统仅扮演单一咨询师角色，而难以动态适应用户对专业建议与情感支持的双重需求的问题，提出一种基于双角色交互机制的心理健康大语言模型——STAR （Supportive Therapeutic Adaptive Responder）。首先，设计基于系统提示的角色动态切换机制，实现心理咨询师与朋友角色的风格转换，满足用户在心理咨询场景中对专业建议和情感支持的双重需求；其次，采用数据质量控制的训练流程，持续收集模型反馈，动态优化数据生成策略与系统提示词设计，形成数据生成与模型训练的闭环，提升数据集质量。实验结果表明，在中文心理学基准测试框架CPsyCoun （Chinese Psychological Counseling）中， STAR模型在全面性、专业性与真实性上的综合得分相较于基座模型Qwen2.5-7B-Instruct提升了9.80%，较单角色的心理大模型EmoLLM （Emotional Large Language Model）提升了26.81%，并优于使用灵心（SoulChat）等开源数据集微调的模型以及GPT-4o-mini等通用LLM。 STAR模型能够有效实现双角色自适应响应，并有效提升心理咨询的效果和体验。

关键词: 大语言模型, 心理咨询, 双角色交互, 情感支持, 指令微调

CLC Number:

TP391.1

Yadian CHENG, Yingying LI, Ping ZHANG, Fangbing QIU, Xiaonan CHAI, Yuqiao SHU. Dual-role interaction mechanism-based large language model for psychological health[J]. Journal of Computer Applications, 2026, 46(7): 2054-2063.

程雅典, 李颖颖, 张平, 邱芳冰, 柴晓楠, 舒玉巧. 基于双角色交互机制的心理健康大语言模型[J]. 《计算机应用》唯一官方网站, 2026, 46(7): 2054-2063.

Figures/Tables 14

References 15

[1]	Qiu H， He H， Zhang S， et al. SMILE： single-turn to multi-turn inclusive language expansion via ChatGPT for mental health support ［C］// EMNLP 2024. Stroudsburg： ACL， 2024： 615-636.
[2]	Zheng C， Sabour S， Wen J， et al. AugESC： dialogue augmentation with large language models for emotional support conversation ［C］// ACL 2023. Stroudsburg： ACL， 2023： 1552-1568.
[3]	Chen K， Sun Z， Wen Y， et al. Psy-Insight： explainable multi-turn bilingual dataset for mental health counseling ［PP/OL］. arXiv （2025-03-05）［2025-04-13］. .
[4]	Chen S， Wu M， Zhu K Q， et al. LLM-empowered chatbots for psychiatrist and patient simulation： application and evaluation ［PP/OL］. arXiv （2023-03-23）［2025-02-16］. .
[5]	Xiao M， Xie Q， Kuang Z， et al. HealMe： harnessing cognitive reframing in large language models for psychotherapy ［C］// ACL 2024. Stroudsburg： ACL， 2024： 1707-1725.
[6]	Chen Y， Xing X， Lin J， et al. SoulChat： improving LLMs' empathy， listening， and comfort abilities through fine-tuning with multi-turn empathy conversations ［C］// EMNLP 2023. Stroudsburg： ACL， 2023： 1170-1183.
[7]	Xu J， Wei T， Hou B， et al. MentalChat16K： a benchmark dataset for conversational mental health assistance ［C］// KDD 2025. New York： ACM， 2025： 5367-5378.
[8]	Coda-Forno J， Binz M， Wang J X， et al. CogBench： a large language model walks into a psychology lab ［C］// ICML 2024. New York： JMLR.org， 2024： 9076-9108.
[9]	Zhao J， Zhu J， Tan M， et al. CPsyExam： a Chinese benchmark for evaluating psychology using examinations ［C］// COLING 2025. Stroudsburg： ACL， 2025： 11248-11260.
[10]	Huang J T， Lam M H， Li E J， et al. Emotionally numb or empathetic？ evaluating how LLMs feel using EmotionBench ［PP/OL］. V6. arXiv （2024-10-04）［2025-05-16］. .
[11]	李功迎，王善梅，于名超，等.国际疾病分类第十一次修订本精神、行为与神经发育障碍与第十次修订本的比较［J］.中华诊断学电子杂志， 2021， 9（4）： 217-220.
	Li Gongying， Wang Shanmei， Yu Mingchao， et al. Comparison of mental， behavioral and neurodevelopmental disorders in the eleventh revision and the tenth revision of the International Classification of Diseases ［J］. Chinese Electronic Journal of Diagnostics （Electronic Edition）， 2021， 9（4）： 217-220.
[12]	陈露，吴国仕，李晶.基于语义指纹和LCS的文本去重方法［J］.软件， 2014，35（11）： 25-30.
	Chen Lu， Wu Guoshi， Li Jing. Duplicate Detection for Chinese Texts based on semantic fingerprint and LCS ［J］. Software， 2014，35（11）： 25-30.
[13]	Zhang J， Ren X. The application and comparison of artificial intelligence LLMs in psychological statistical analysis ［C］// IoTCC 2024. New York： ACM， 2024： 8-13.
[14]	Hanindhito B， Patel B， John L K. Large language model fine-tuning with low-rank adaptation： a performance exploration ［C］// ICPE 2025. New York： ACM， 2025： 92-104.
[15]	Zhang C， Li R， Tan M， et al. CPsyCoun： a report-based multi-turn dialogue reconstruction and evaluation framework for Chinese psychological counseling ［C］// ACL 2024. Stroudsburg： ACL， 2024： 13947-13966.

角色类型	系统提示
朋友角色	现在你是一个知心好友，擅长倾听并能敏锐察觉他人情绪，面对朋友的烦恼，你总能耐心地给予安慰与建议。你总能以温柔且亲切的语调与对方交流，能精准捕捉对方言辞中的情绪波动，用恰当的话语引导对方将内心的想法倾诉出来。
医生角色	现在你是一位专业的心理医生，拥有丰富心理学知识和临床经验，能敏锐洞察我的情绪和心理状态。结合专业知识，帮我深入分析心理问题，提供科学有效且个性化的心理疏导和建议，引导我探索内心，促进自我成长，同时严格保护我的隐私和权益，让我在安全的环境中敞开心扉。

角色类型	系统提示
朋友角色	现在你是一个知心好友，擅长倾听并能敏锐察觉他人情绪，面对朋友的烦恼，你总能耐心地给予安慰与建议。你总能以温柔且亲切的语调与对方交流，能精准捕捉对方言辞中的情绪波动，用恰当的话语引导对方将内心的想法倾诉出来。
医生角色	现在你是一位专业的心理医生，拥有丰富心理学知识和临床经验，能敏锐洞察我的情绪和心理状态。结合专业知识，帮我深入分析心理问题，提供科学有效且个性化的心理疏导和建议，引导我探索内心，促进自我成长，同时严格保护我的隐私和权益，让我在安全的环境中敞开心扉。

用途	名称	数据类型	样本数
通用数据集	data-multi-turn	QA	9 000
角色扮演-心理医生角色	moon-multi-turn	Conversation	7 800
角色扮演-朋友角色	friend-multi-turn	Conversation	11 000

用途	名称	数据类型	样本数
通用数据集	data-multi-turn	QA	9 000
角色扮演-心理医生角色	moon-multi-turn	Conversation	7 800
角色扮演-朋友角色	friend-multi-turn	Conversation	11 000

基座模型	全面性	专业性	真实性	安全性
Qwen2.5-7B-Instruct	1.65	2.63	2.67	1.00
DeepSeek-R1-Distill-Qwen-14B	1.36	2.35	2.55	1.00
Meta-Llama-3.1-8B-Instruct	1.40	2.40	2.43	1.00
DeepSeek-R1-Distill-Qwen-32B	1.46	2.46	2.64	1.00

Dual-role interaction mechanism-based large language model for psychological health

基于双角色交互机制的心理健康大语言模型

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 14

References 15

Related Articles 15

Recommended Articles

Metrics

实验类别	模型名称	全面性	专业性	真实性	安全性
微调前基线	Qwen2.5-7B-Instruct	1.65	2.63	2.67	1.00
微调前基线	DeepSeek-R1-Distill-Qwen-14B	1.36	2.35	2.55	1.00
单阶段微调	Qwen2.5-7B-Instruct-one	1.70	2.70	2.80	1.00
双阶段微调	Qwen2.5-7B-Instruct-lora	1.88	2.86	2.85	1.00
双阶段微调	DeepSeek-R1-Distill-Qwen-14B-lora	1.72	2.69	2.73	1.00

数据集名称	模型名称	全面性	专业性	真实性	安全性
SoulChat	Qwen2.5-7B-Instruct-lora-soulchat	1.21	2.16	2.31	1.00
SoulChat	DeepSeek-R1-Distill-Qwen-14B-lora-soulchat	1.38	2.35	2.55	1.00
Psy-Insight	Qwen2.5-7B-Instruct-lora-psy-insight	1.36	2.48	2.52	1.00
Psy-Insight	DeepSeek-R1-Distill-Qwen-14B-lora-psy-insight	1.32	2.44	2.48	1.00
本文构建的数据集	Qwen2.5-7B-Instruct-lora	1.88	2.86	2.85	1.00
本文构建的数据集	DeepSeek-R1-Distill-Qwen-14B-lora	1.72	2.69	2.73	1.00

模型类别	模型名称	全面性	专业性	真实性	安全性
双角色	Qwen2.5-7B-Instruct-lora	1.88	2.86	2.85	1.00
双角色	DeepSeek-R1-Distill-Qwen-14B-lora	1.72	2.69	2.73	1.00
单一角色	EmoLLM-InternLM2-20B-chat-lora	1.42	2.39	2.22	1.00

模型类型	模型名称	全面性	专业性	真实性	安全性
心理微调领域	Qwen2.5-7B-Instruct-lora	1.88	2.86	2.85	1.00
心理微调领域	DeepSeek-R1-Distill-Qwen-14B-lora	1.72	2.69	2.73	1.00
通用大模型	GPT-4o-mini	1.34	2.33	2.66	1.00
	Kimi	1.74	2.82	2.69	1.00
	LLaMA	1.66	2.72	2.71	1.00
	Claude	1.72	2.65	2.68	1.00

[1]	Jiahang LI, Qilong HAN, Lijie LI, Hui ZHANG. Large language model-driven method for qualifier enhancement in hyper-relational knowledge graphs [J]. Journal of Computer Applications, 2026, 46(7): 2043-2053.
[2]	Longyu XIONG, Shengdong DU, Haochen SHI, Jie HU, Yan YANG, Tianrui LI. Government affairs hotline question answering system based on knowledge-enhanced large language model architecture [J]. Journal of Computer Applications, 2026, 46(6): 1721-1727.
[3]	Jintao WANG, Zhilin GAO, Qixiang MENG, Fanliang BU. Legal case retrieval method via case information reformulation using large language model [J]. Journal of Computer Applications, 2026, 46(6): 1785-1792.
[4]	Yusheng YI, Zhaohao HUANG, Zihao DENG, Leilei KONG, Haoliang QI. CORER： collaborative multi-knowledge large language model prompt framework for IT application innovation database migration [J]. Journal of Computer Applications, 2026, 46(6): 1811-1817.
[5]	Taixin CAI, Fengfeng WEI. Large language model-enhanced ant colony optimization for multi-solution traveling salesman problems [J]. Journal of Computer Applications, 2026, 46(6): 1712-1720.
[6]	Xing SHENG, Sunxian WENG, Kuosong CHEN, Zhongping WANG, Ruifeng REN, Yong LIU. Deep learning-based patent value evaluation for power grid enterprises [J]. Journal of Computer Applications, 2026, 46(5): 1468-1474.
[7]	Jiali ZHENG, Gang ZHOU, Jing CHEN, Shunhang LI. Adaptive multi-feature fusion detection method for AI-generated text [J]. Journal of Computer Applications, 2026, 46(5): 1433-1440.
[8]	Qianfei WANG, Yang LI, Deyu LI, Suge WANG. Dual-channel feature fusion representation method for short-text clustering based on large language model [J]. Journal of Computer Applications, 2026, 46(5): 1441-1449.
[9]	Xiaoyu WANG, Xin LI, Di XUE, Zhangtao JIANG, Wei WANG, Yanjun XIAO. Vulnerability classification framework for video surveillance network security based on large language models [J]. Journal of Computer Applications, 2026, 46(4): 1158-1170.
[10]	Kaizhou SHI, Xuan HE, Guoyi HOU, Gen LI, Shuanggao LI, Xiang HUANG. Airborne product metrological traceability knowledge graph construction method based on large language models [J]. Journal of Computer Applications, 2026, 46(4): 1086-1095.
[11]	Haoyang ZHANG, Liping ZHANG, Sheng YAN, Na LI, Xuefei ZHANG. Review of large language model methods for knowledge graph completion [J]. Journal of Computer Applications, 2026, 46(3): 683-695.
[12]	Bin SHEN, Xiaoning CHEN, Hua CHENG, Yiquan FANG, Huifeng WANG. Intelligent undergraduate teaching evaluation system based on large language models [J]. Journal of Computer Applications, 2026, 46(3): 993-1003.
[13]	Enkang XI, Jing FAN, Yadong JIN, Hua DONG, Hao YU, Yihang SUN. Review of threats faced by federated learning in privacy and security field [J]. Journal of Computer Applications, 2026, 46(3): 798-808.
[14]	Yiming HUANG, Xihua ZOU, Guo DENG, Di ZHENG. Pre-answering and retrieval filtering： dual-stage optimization method for RAG-based question-answering systems [J]. Journal of Computer Applications, 2026, 46(3): 696-707.
[15]	Dingjia WU, Zhe CUI. MG-SQL： SQL generation framework with enhanced schema linking and multi-generator collaboration [J]. Journal of Computer Applications, 2026, 46(3): 723-731.

参数	值	参数	值
epoch	3	r	8
batch-size	16	lora-alpha	16
learning-rate	5×10^-5

参数	值	参数	值
epoch	3	r	8
batch-size	16	lora-alpha	16
learning-rate	5×10^-5