Deep learning-based hybrid model for knowledge tracing and personalised e-learning

Personalised learning has emerged as an essential part of modern education systems, demanding intelligent models that can adapt to each learner's developing knowledge and preferences. Our paper proposes a novel hybrid deep learning framework that utilises sequential modelling using Long Short-Term Memory-based Deep Knowledge Tracing (LSTM-DKT), SelfAttentive Interaction Network for Knowledge Tracing (SAINT), and Variational Autoencoder-based Collaborative Filtering (VAECF) to produce personalised learning recommendations. The model combines the strengths of both LSTM and SAINT to predict a learner's mastery over time, and integrates these predictions into the VAE-CF model to align the learner's knowledge with their latent content preferences. The integrated architecture employs learning patterns and collaborative filtering to address challenges related to cold start and sparsity in educational data. This hybrid system outperformed the baseline models in both mastery prediction and personalized recommendation tasks. The research shows that cognitive tracing and collaborative filtering significantly enhance the accuracy and relevance of personalised content, providing a strong basis for the development of more flexible and student-centered e-learning platforms.