CITE: A Classroom Interaction and Teaching Evaluation Algorithm Based on BERT and GAT

Abstract

Traditional classrooms often face challenges in monitoring student engagement and providing timely feedback, especially in dynamic teaching environments. This study focuses on primary school science education classes, targeting fifth-grade students, and proposes an intelligent classroom evaluation method that integrates BERT and Graph Attention Network (GAT). BERT captures global semantic features from classroom discourse, while GAT models the interaction structure between teachers and students. Real data from fifth-grade science education classes were collected to validate the model, and visualization experiments were conducted on the results. The experimental result shows that the proposed method outperforms traditional approaches in terms of both accuracy and responsiveness. Furthermore, the visualized output helps teachers intuitively grasp students’ engagement levels. These results demonstrate the effectiveness of the model in enabling real-time, data-driven feedback and supporting adaptive teaching strategies in primary education.