Integrating BiLSTM-based Sentiment Analysis and GCN for Academic Performance Prediction under Academic Stress

Abstract

In order to build a more comprehensive and refined emotional perception mechanism, future intelligent education systems urgently need to integrate multidimensional emotional modeling capabilities. In response to this demand, this article constructs a bidirectional long short-term memory network (BiLSTM) emotion classification model, which is based on text data posted by students on social media platforms and integrates self attention mechanisms. The study selected online course learning data and social media data from 3146 college students at a certain university. The accuracy comparison between the model proposed in this article and the model used in the task of identifying students' mental health status. The experimental results show that the model proposed in this paper is significantly better than the comparative models in terms of accuracy indicators. The BiLSTM structure based on self attention mechanism has been introduced, which can more effectively capture fine-grained emotional features in social media texts. The combination of BiLSTM and self attention mechanism effectively captures emotional and psychological state features in student texts, with an accuracy performance of over 90%. GCN enhances situational awareness prediction ability by modeling student social relationship maps, integrating group behavior and individual learning data. The synergistic effect of the two enables the model to maintain high accuracy while also having good recall ability, thereby achieving more robust and balanced academic performance prediction.