International Open Access Journal Platform

基于行为数据和ERPs 数据的社交焦虑诊断——人工智能方法的应用

Diagnosis of Social Anxiety Based on Behavioral Data and ERPs: Application of Artificial Intelligence Methods

Psychology of China / 2025,7(2)：221-229 / 2025-03-04 112 71

View Download PDF

• Authors: 马智泽     
• Information:

  宁夏大学教师教育学院，银川

• Keywords: 执行功能；社交焦虑；人工智能方法；ERPs；多模态数据融合
• Executive function; Social anxiety; Artificial intelligence methods; ERPs; Multimodal data fusion
• Abstract: 社交焦虑是大学生中普遍存在的心理问题，会严重损害他们的心理健康和执行功能。本研究采用执行功能相关的三个实验范式：Go/No-go（抑制功能）、任务转换（认知灵活性）、N-back（工作记忆），收集大学生的行为数据和脑电数据，并应用人工智能方法（机器学习和深度学习技术），探索对社交焦虑个体的客观识别方法。结果显示，在仅使用行为数据进行分类时，深度学习模型的表现优于传统、集成机器学习模型（如支持向量机、随机森林）。同样，在使用脑电数据进行分类时，深度学习模型也展现出显著的优势。特别地，基于行为和脑电数据的多模态深度学习模型（BEPTCNN）在各项评估指标上都取得了最优的表现。这表明，多模态数据融合能够提供更丰富的信息，有助于提高社交焦虑的诊断准确性。本研究结果为发展基于人工智能的社交焦虑诊断工具提供了有力支持，为进一步揭示社交焦虑的生理机制和改善干预方案奠定了基础。未来研究可以进一步探索深度学习模型的泛化性和可解释性，以及将其应用于实时诊断系统的可能性。
• Social anxiety is a prevalent psychological issue among college students, severely impairing their mental health and executive functioning. This study employed three executive function-related experimental paradigms—Go/No-go (inhibitory control), task-switching (cognitive flexibility), and N-back (working memory)—to collect behavioral and electroencephalogram (EEG) data from college students. Artificial intelligence methods, including machine learning and deep learning techniques, were applied to explore objective identification approaches for individuals with social anxiety. Results demonstrated that when using only behavioral data for classification, deep learning models outperformed traditional and ensemble machine learning models (e.g., Support Vector Machine, Random Forest). Similarly, deep learning models exhibited significant advantages in EEG data classification. Notably, the multimodal deep learning model (BEPTCNN) integrating behavioral and EEG data achieved optimal performance across all evaluation metrics. This indicates that multimodal data fusion provides richer information, enhancing diagnostic accuracy for social anxiety. The findings strongly support the development of AI-based diagnostic tools for social anxiety and lay a foundation for further elucidating its physiological mechanisms and improving intervention strategies. Future research should explore the generalizability and interpretability of deep learning models, as well as their potential application in real-time diagnostic systems.
• DOI: https://doi.org/10.35534/pc.0702035
• Cite: 马智泽．基于行为数据和 ERPs 数据的社交焦虑诊断 ——人工智能方法的应用［J］．中国心理学前沿，2025，7（2）：221-229．