Construction of a College Student Psychological Crisis Early Warning System Based on Big Data and Knowledge Management

Abstract

College students' psychological crises are concealed and diverse, making traditional questionnaire and interview screenings limited by low frequency and time lag. This study constructs a big data-driven closed-loop early warning system that fuses online learning behavior, social semantics, and wearable physiological signals. To address cross-modal heterogeneity, the authors use a multi-task deep learning architecture with Text-CNN, Bi-LSTM, and GCN to capture semantic, emotional, and temporal features. Dynamic risk classification and reinforcement learning scheduling enable real-time interpretable early warning, while differential privacy and federated learning ensure data security. Two-semester real-world deployment shows improved sensitivity and reduced false positives. The system integrates with campus platforms to provide customized intervention strategies, and experiments confirm low latency and energy consumption compared with baselines. This research provides a scalable paradigm for digital campus mental health governance.