Application of Social Network Text Sentiment Analysis in Monitoring Students’ Mental State

Abstract

Social-network text provides real-time and natural signals for monitoring students’ emotional fluctuations, but its informal, implicit, and context-dependent expressions make reliable psychological-state assessment difficult. This study develops a deep-learning-based sentiment analysis framework for student mental-state monitoring by integrating contextual semantic representation and temporal emotion modeling. Social-network texts are filtered, cleaned, semantically normalized, and vectorized after processing emoticons, abbreviations, and nonstandard expressions. A BERT pre-trained model is used to extract contextual semantic features, and a BiLSTM network captures sequential emotional dependencies and semantic reversal patterns. An emotion-intensity attention mechanism further enhances text segments associated with stress, anxiety, and depressive tendencies. Based on the sentiment distribution within continuous time windows, a psychological-state scoring model and a risk-discrimination function combining emotion intensity and temporal variation are constructed. Experimental results on 12,500 annotated texts from 850 users show an average F1 score of 0.89 for multi-class emotion recognition, with depressive emotion recognition reaching 0.91. High-risk student identification achieves an accuracy of 0.94. The method provides a text-signal processing and temporal risk-assessment framework for intelligent monitoring and early warning systems.