Research on Employment Psychology and Anxiety and Depression of College Students Under Deep Learning

Faced with the multiple pressures of employment competition, economic fluctuation, and insufficient psychological support, this study builds a Convolutional Neural Network (CNN)–Bidirectional Long Short-Term Memory (BiLSTM)–Attention dual-channel model based on the multimodal longitudinal data of fresh graduates from eight universities to jointly predict anxiety, depression, and cognitive status under employment pressure. The model performance (AUC=0.92, F1=0.88, Kappa=0.71) is significantly better than the baseline. Explanatory analysis identified economic pressure, screen length, and delivery frequency as the main risk factors, and social support, sleep, and practice had protective effects. After landing, 312 cases of early intervention were realized, and the average values of anxiety and depression decreased by 0.19 and 3.8, respectively, and the satisfaction rate was greater than 90%. The research puts forward an integrated paradigm of multimodal modeling + double interpretability, which provides data-driven quantitative support for psychological intervention and policymaking in colleges and universities.