基于聊天机器人的心理干预已成为提升大学生心理健康的有效途径。聊天机器人的拟人设计通过模仿人类语言和情感反应，有助于与用户建立关系，促进心理支持的接受度。过往研究主要集中在聊天机器人的任务执行能力上，忽略了拟人元素对用户体验和治疗结果的影响。因此，本研究通过结合积极心理干预和检索增强生成（RAG）等人工智能技术，设计并实现了四种不同拟人角色（伙伴、专家）及自我表露（情绪、事实）的聊天机器人组合，以探索拟人设计对积极心理干预效果的影响及其对使用意愿的作用机制。

研究一在线招募了105名大学生及研究生，采用2×2的被试间实验设计，验证拟人身份及自我表露对聊天机器人干预效果的影响。干预前后结果表明，专家与事实表露组合的干预效果最佳，其在生活满意度（p = 0.016，Cohen’s d

 = 0.494）、积极情绪（p = 0.006，Cohen’s d = 0.571）、心理韧性（p < 0.001，       Cohen’s d = 0.684）及积极心理资本（p < 0.001，Cohen’s d = 1.114）上均显著提升，同时在降低消极情绪（p < 0.001，Cohen’s d = 0.719）方面表现显著。

研究二在研究一的基础上进一步探索聊天机器人的拟人设计及性别分组对用户心理感知的影响，以及心理感知如何影响数字治疗联盟（DTA）的建立和使用意愿。基于理论框架构建研究模型，通过偏最小二乘法路径模型得到验证。结果发现，女性被试更易感知聊天机器人具备高的心智水平（p = 0.002， B = - 0.295）、建立信任（p < 0.001，B = - 0.482）及亲密关系（p < 0.001，B =   - 0.358）；被试更易感知事实聊天机器人有较高心智（p = 0.004，B=-0.265）和建立信任（p = 0.019，B = - 0.204）。不同拟人角色在用户感知层面未显著差异。心智感知（p = 0.017，B = 0.230）、信任（p = 0.003，B = 0.287）及亲密 （p = 0.004，B = 0.285）均显著影响数字治疗联盟的建立，进而显著影响用户的使用意愿（p < 0.001，B = 0.751），从而验证了研究模型的有效性。

本研究强调了拟人化设计在提升聊天机器人干预效果和用户使用意愿中的关键作用。通过深入分析，揭示了拟人化设计如何显著影响用户心理感知和数字治疗联盟的建立，为心理健康领域的聊天机器人设计供了新的视角和策略。

Psychological interventions based on chatbots have become an effective way to enhance the mental health of college students. The anthropomorphic design of chatbots, by mimicking human language and emotional responses, helps establish relationships with users and promotes the acceptance of psychological support. Past research has primarily focused on the task execution capabilities of chatbots, overlooking the impact of anthropomorphic elements on user experience and therapeutic outcomes. Therefore, this study combines positive psychological interventions and artificial intelligence technologies like Retrieval-Augmented Generation (RAG) to design and implement chatbots with four different anthropomorphic roles (companion, expert) and self-disclosure modes (emotional, factual), exploring the effects of anthropomorphic design on the outcomes of positive psychological interventions and its mechanism on user willingness.

Study one recruited 105 college and graduate students online and used a 2x2 between-subjects experimental design to verify the effects of anthropomorphic identity and self-disclosure on the outcomes of chatbot interventions. The results before and after the intervention indicate that the combination of expert role and factual disclosure had the best intervention outcomes, showing significant improvements in life satisfaction (p = 0.016, Cohen’s d = 0.494), positive emotions (p = 0.006, Cohen’s d = 0.571), psychological resilience (p < 0.001, Cohen’s d = 0.684), and positive psychological capital (p < 0.001, Cohen’s d = 1.114), while also significantly reducing negative emotions (p < 0.001, Cohen’s d = 0.719).

Building on the first study, Study two further explores the impact of chatbot anthropomorphic design and gender grouping on users' psychological perceptions and how these perceptions affect the establishment of a digital therapeutic alliance (DTA) and user willingness. The research model was validated through a Partial Least Squares Path Modeling. The results revealed that female participants were more likely to perceive high levels of mind (p = 0.002, B = - 0.295), trust (p < 0.001, B = - 0.482), and intimacy (p < 0.001, B = - 0.358) in the chatbots; factual disclosure chatbots made it easier for participants to perceive a higher level of mind (p = 0.004, B = - 0.265) and trust (p = 0.019, B = - 0.204). There were no significant differences in user perceptions among different anthropomorphic roles. Perceived mind (p = 0.017, B = 0.230), trust (p = 0.003, B = 0.287), and intimacy (p = 0.004, B = 0.285) all significantly influenced the establishment of a digital therapeutic alliance, which in turn significantly affected user willingness (p < 0.001, B = 0.751), thereby validating the effectiveness of the research model.

This research emphasizes the crucial role of anthropomorphic design in enhancing the effects of chatbot interventions and user willingness. Through in-depth analysis, it reveals how anthropomorphic design significantly affects users' psychological perceptions and the establishment of a digital therapeutic alliance, providing new perspectives and strategies for the design of chatbots in the field of mental health.