Research on Identifying Psychological Health Risks of College Students Based on Knowledge Graph and Multimodal Text Analysis

Abstract

This research constructs a “five-in-one” framework—comprising knowledge graph construction, multimodal data preprocessing, feature fusion, risk identification, and empirical verification— to enable the early detection of mental health risks among college students in intelligent campus environments. By integrating psychological, academic, behavioral and environmental information with advanced modeling, the study supports more accurate risk monitoring. Firstly, the core influencing factors of college students’ mental health risks are systematically sorted out. Based on social ecology theory and stress coping theory, a knowledge graph of mental health risks covering five dimensions: individual psychology, academic development, social support, life behavior, and environmental adaptation is constructed; Secondly, collect multimodal text data from college students (social media texts, course reviews, psychological counseling records, voice logs), and construct a multimodal text dataset through preprocessing steps such as text cleaning, emotional annotation, and feature extraction; Once again, design a multidimensional feature system that integrates text semantic features, emotional features, behavioral features, and knowledge graph correlation features, and construct a mental health risk identification model based on deep learning models (BERT+BiLSTM).