Research on the Identification and Intervention Mechanism of College Students’ Employment Psychological Stress by Integrating Multi source Text Data and Deep Learning

Abstract

Against the background of increasing graduate numbers, structural employment contradictions, and occupational uncertainty, employment-related psychological stress among college students has become a major issue affecting mental health and employment quality. Traditional identification methods based on questionnaires and interviews suffer from lag, subjectivity, limited coverage, and privacy concerns. This study introduces multi-source text data and deep learning into employment-stress recognition and constructs a full-process mechanism of multi-source collection, deep recognition, dynamic warning, precise intervention, and effect feedback. High-frequency campus messages, employment consultation texts, questionnaires, and reflective job-search narratives are integrated to capture stress cues related to anxiety, frustration, decision difficulty, avoidance, and social comparison. A deep learning model combining BERT/RoBERTa, TextCNN, BiLSTM, attention mechanisms, and decision fusion is proposed to identify stress levels, stress types, emotional tendencies, and risk levels. The framework supports quasi-real-time monitoring and personalized intervention, providing an intelligent signal-fusion approach for campus mental-health and employment guidance systems.