A Multi-Component Deep Learning Framework for Psychological Profiling of College Students Using Behavioral and Sentiment Data

Abstract

This study constructs a psychological portrait model of college students based on deep learning, aiming to accurately depict the psychological characteristics of college students. Through the analysis of the college students' mental health dataset, the performance of the proposed model is compared with that of the support vector machine, naive Bayes, decision tree, multi-layer perceptron and baseline model. It specifies that the psychological profiling model integrates a bidirectional Long Short-Term Memory (BiLSTM) network for sentiment analysis, Deep Reinforcement Learning (DRL) for behavior pattern extraction, and a standard LSTM architecture for dynamic psychological state prediction. The abstract also notes that the dataset comprises multi-dimensional data on college students’ mental health, including emotional states, academic pressure, and social interaction patterns. Preprocessing involved text vectorization through word embedding and normalization of behavioral features. Model configurations, such as the use of high-dimensional embeddings and multi-layer network training, are briefly referenced to highlight the technical depth of the architecture. These additions provide critical context for understanding the modeling pipeline and support future reproducibility. In terms of accuracy, the comprehensive accuracy of the proposed model reaches 0.90, which is much higher than 0.75 of the support vectors machines, 0.68 of the naive Bayes, 0.73 of the decision trees, 0.78 of the multi-layer perceptron and 0.58 of the baseline models. In terms of recall rate, the comprehensive recall rate of the proposed model is 0.88, which is also ahead of other models. Indicators such as F1 value and root mean square error also show the advantages of the proposed model. The experimental results show that the model performs well in predicting multi-dimensional psychological characteristics such as emotional tendency, academic anxiety, social status, and emotional fluctuations, and can effectively capture the changes in the psychological state of college students. However, the model has certain limitations in external validity and generalizability. In the future, it is necessary to expand the scope of the dataset and optimize the model structure to improve its performance.