Deep learning-based psychoeducational knowledge tracking model in the big data perspective

Under the background of big data, it is increasingly urgent for the field of psychological education to accurately track students’ knowledge. Aiming at the issue that existing studies ignore the internal correlation between exercises and knowledge points, this paper firstly improves Long Short-Term Memory network (MELSTM) based on memory extension, and then constructed the psychoeducational Knowledge Graph (KG). Word2Vec and bidirectional MELSTM were used to convert the exercise response sequences into low-dimensional dense vectors, and KG embedding representation was carried out by TransR model. The influence degree of precursor knowledge on prediction results was explored through attention mechanism. Finally, the prediction results are obtained through the fully connected network. Experiments on three education datasets show that the AUC values of the proposed model are improved by at least 8.6%, 3.4% and 5.5%, which can track students’ knowledge status more accurately.