An Intelligent Intervention System for Adolescent Anxiety Based on Multimodal Deep Learning and Adaptive Strategies
作者
Lie Ma,Yanglong Chen,Shengnan Qiao,Chen Shao,R.Y. Zhang,Yuxuan Liang,Xuanming Si
标识
DOI:10.1145/3784013.3784031
摘要
Addressing the prominent contradiction between the high incidence rate and low intervention rate of adolescent anxiety disorders, this paper designs and implements an intelligent intervention system based on multimodal deep learning and contextual reinforcement learning. The system achieves accurate identification of anxiety states by collaboratively analyzing users' social media text and physiological signals from wearable devices (heart rate, electrodermal activity), and dynamically generates personalized intervention strategies based on this. This study constructed a multimodal dataset containing 5,012 adolescents (CAS-5012). In closed testing, the system achieved an accuracy of 93.5% and an F1-score of 92.8% for anxiety recognition. A four-week randomized controlled trial showed that the reduction in anxiety scale (GAD-7) scores was significantly greater in the intervention group (n=50) than in the control group (p = 0.003), with an effect size (Cohen's d) of 1.24. This study confirms the effectiveness of the proposed system in the early identification and intervention of adolescent anxiety disorders, providing empirical evidence for the application of computational psychiatry in real-world scenarios.