社交焦虑障碍是当代大学生群体中普遍存在的心理问题。它可能会对个体的心理健康和生活质量产生负面影响。然而，传统社交焦虑障碍诊断与干预存在一定的局限性：患者可能会像逃避其他社交一样逃避医疗服务，传统的诊断方法依赖主观评分，缺乏生理数据的客观评价。本研究基于认知行为理论与游戏化方法，结合虚拟现实与多通道生理数据反馈技术，构建一个自助式社交焦虑倾向检测严肃游戏《新生日记》。具体包括：第一，在虚拟现实暴露疗法与认知行为理论指导下，设计了场景关卡与故事剧本，实现了虚拟现实中手柄、眼动、语音的多通道交互与数据埋点方案；第二，按照设计方案开发了游戏系统，以模块化控制的系统架构增强游戏的扩展性与健壮性，并明确了多通道数据的采集规范，为后续研究输出客观数据；第三，开展线下实验并收集评价量表与游戏数据，使用 XGBoost 算法构建了基于玩家游戏数据的社交焦虑倾向检测模型。通过实验数据分析得到以下结论：在游戏内容层面，游戏体验（iGEQ， ? =7.30）和系统易用性（SUS， ? = 8.25）得到了玩家的较高认可，通过 ? 检验可知患者社交焦虑倾向对游戏焦虑感知程度有显著影响关系（? = 0.001）。在模型效果层面，构建的逻辑回归、线性回归、多分类机器学习模型的 F1 值都达到 0.89，其中线性回归模型的检测效果最佳（F1 = 0.93），证明研究构建的检测模型能有效识别玩家的社交焦虑倾向。本研究创新点主要体现在两方面：一是内容层面，基于游戏化方法将认知行为训练任务融入剧本与关卡设计中，通过多分支叙事的方法塑造完整、个性、动态的游戏内容体验；二是数据层面，将多通道数据（手柄、眼动、语音）采集自然地融入游戏交互设计中，将玩家主观行为与生理反应有效地转化为客观数据的分析。
Social anxiety disorder is a prevalent psychological issue among college students that has negative effects on mental health and well-being. However, traditional diagnosis and treatment have limitations. Patients may avoid seeking medical assistance, just as they avoid other social situations. Moreover, conventional diagnostic methods rely on subjective questionnaires without objective analysis of physiological data.This study investigates how to create a self-help serious game “Freshman Diary” for detecting social anxiety using virtual reality and multi-channel physiological data technology. Firstly, this study used gamification methods and guidance from cognitive behavioral psychology theory to design scene levels and storylines. And this project established a multi-channel interaction and data embedding scheme that utilizes hand-held controllers, eye movement, and speech. Secondly, the modular control system architecture was developed to enhance the game’s scalability and robustness. Standards of multi-channel data collection were also established to provide objective data support for future research. Thirdly, this research conducted offline experiments to collect questionnaires and game data. Then, the research developed a social anxiety tendency detection model using the XGBoost algorithm.Data analysis revealed that players highly recognized player experience and system usability (iGEQ, ? = 7.30 and SUS, ? = 8.25, respectively). T-test results showed students with higher social anxiety tendencies experienced more anxiety in this game (? = 0.001). The F1-Score of constructed detection models were all above 0.89, with the linear regression model having the best detection effect ( F1 = 0.93), proving the model’s effectiveness in identifying players’ social anxiety tendencies.This study has two key innovations. It integrates cognitive-behavioral training tasks into the game’s design, offering a personalized experience through multi-branch storytelling. Additionally, this study seamlessly incorporates various types of data (controller, eye tracking, speech volume) into game interaction design. This enables analysis of player behavior and physiological responses.
