Sentiment Analysis of Mental Health among Chinese College Students Using Hybrid Modeling

Abstract

As mental health issues among Chinese college students have evolved into a significant social concern within the higher education domain, this study addresses the urgent need for mental health monitoring by proposing a sentiment analysis technical solution that combines deep learning with traditional machine learning. Using "college students" as the scenario keyword, 4,201 relevant posts were collected from the Sina Weibo platform between 2023 and 2025. After data cleaning, word segmentation, and annotation, a tri-classification dataset containing 3,950 effectively labeled entries was constructed. The study first employed the TF-IDF method to extract text features, revealing that academic stress-related vocabulary had the highest weights, reflecting that academic burden is the primary psychological stressor. In model evaluation, traditional machine learning performed best with Random Forest (accuracy: 0.792), while the SVM model exhibited overfitting. In contrast, the hybrid deep learning model CNN-BiLSTM-Attention demonstrated comprehensive advantages (accuracy: 0.813, F1-score: 0.829), particularly excelling in identifying neutral and negative sentiments. Its training and testing losses were also significantly lower than those of other machine learning models. Therefore, in complex contexts, deep learning models achieve higher recognition accuracy than traditional machine learning models. Finally, the study provides a technical solution for college mental health monitoring that balances timeliness and accuracy, aiming to offer precise, real-time, and compliant support for university psychological service systems.