Data Collection and Analysis of Psychological Health Signs of College Students Based on Artificial Intelligence Technology

Abstract

The problem of mental health concerns among university students is growing more and more noticeable, and early identification of mental health signs is crucial for intervention. Current assessment methods rely on questionnaire data, which suffer from inefficiencies in data statistics. Taking Rednote as a case study, this research employs web crawling technology to gather diverse user information. By utilizing data mining techniques, linguistic, emotional, and behavioral features are extracted, and a depression knowledge graph is constructed to model the deep-level associations among these features. Additionally, the study selects a Convolutional Neural Network (CNN) and a Gated Recurrent Unit (GRU) to develop a predictive model that integrates textual features with user characteristics, ultimately yielding mental health predictions. The results indicate that the predictive model achieves an accuracy of 87.3% and an F1 score of 86.5%. Without the knowledge graph, the accuracy drops to 81.2%, representing a 6.1% decrease compared to the predictive model, demonstrating that the knowledge graph can effectively identify key depression-related pathways. The F1 score for the combined CNN and GRU model reaches 98.5%, showing a 6.3% improvement over the GRU alone. This research provides a feasible approach for social media data-driven mental health monitoring, with the application of the knowledge graph enhancing model interpretability and aiding in the development of precise psychological intervention strategies. It offers an automated screening tool for college psychological counseling services and provides data support for depression prevention research in the realm of public health.