Learning style detection based on graph embedding enhancement

In the “Internet and Education” era, accurately detecting students' learning styles is an indispensable part of personalized educational resource recommendations. Learning style is crucial for providing customized knowledge services and implementing effective learning guidance. This study aims to propose a student learning style detection method enhanced by graph embedding through modeling online learning data. Firstly, we analyze students' behavioral data and extract original behavioral features that reflect their learning styles. Then, we introduce a Bipartite Graph Autoencoder (BGAE) model to model student-learning resource interactions and extract graph embedding features. Finally, these features are then combined with original features to improve prediction performance. We conduct comparative experiments on multiple machine learning algorithms on a real dataset and ultimately determined the best performing random forest as the prediction model. The experimental results show that the classification model based on fused graph embedding features is significantly better in accuracy and stability than the model based solely on original features. The accuracy of the random forest is improved to 98.57%, which verifies the effectiveness of our research method and provides certain support for achieving personalized and precise recommendations in educational management.