Depression Detection Based on Multi-source Information Fusion

Abstract

In recent years, depression has severely impacted the physical and mental health and safety of adolescents, making both treatment and prevention of depression in adolescents crucial. To enhance depression detection, this paper proposes a depression detection model based on the RBF kernel and the BERT Transformer. This model extracts features from text data and structured feature data using a text feature extraction module and a structured data extraction module, respectively. After feature fusion, it uses support vector machines for classification and finally calculates the probability of depression and the depression index of the student. Experimental validation demonstrates excellent results on a student depression dataset containing 27,902 records, achieving an AUC of 1.00. The weighted depression dictionary fusion method increases the sensitivity of high-risk semantic features, facilitating depression detection. Compared to traditional models, the RBF kernel and the BERT Transformer provide a superior solution for building depression detection systems, contributing to enhanced mental health protection and self-defense awareness.