Adherence to healthy diets reduces chronic illness risk, yet rates remain low. Large Language Models (LLMs) are increasingly used for health communication but often struggle to engage individuals with ambivalent intentions at a pivotal stage of the Transtheoretical Model (TTM). We developed CounselLLM, an open-source model enhanced through persona design and few-shot, domain-specific prompts grounded in TTM and Motivational Interviewing (MI). In controlled evaluations, CounselLLM showed stronger use of TTM subprocesses and MI affirmations than human counselors, with comparable linguistic robustness but expressed in more concrete terms. A user study then tested CounselLLM in an interactive counseling setting against a baseline system. While knowledge and perceptions did not change, participants' intentions for immediate dietary change increased significantly after interacting with CounselLLM. Participants also rated it as easy to use, understandable, and supportive. These findings suggest theory-driven LLMs can effectively engage ambivalent individuals and provide a scalable approach to digital counseling.
翻译:坚持健康饮食可降低慢性病风险,但实际遵循率仍然较低。大型语言模型(LLMs)在健康沟通中的应用日益增多,但在跨理论模型(TTM)的关键阶段,往往难以有效吸引具有矛盾意向的个体。我们开发了CounselLLM,这是一个基于TTM和动机性访谈(MI)理论,通过角色设计和少量领域特定提示增强的开源模型。在受控评估中,CounselLLM比人类咨询师表现出更强的TTM子过程运用和MI肯定技巧,语言稳健性相当,但表达更为具体。随后,一项用户研究在交互式咨询环境中将CounselLLM与基线系统进行比较。虽然参与者的知识和认知未发生改变,但在与CounselLLM互动后,他们立即改变饮食的意向显著增强。参与者还评价其易于使用、易于理解且具有支持性。这些发现表明,理论驱动的LLMs能有效吸引矛盾意向个体,并为数字化咨询提供可扩展的途径。