Personalized Learning Recommendations with Multi-Layered Neural Collaborative Filtering

Abstract

The emergence of numerous learning platforms through online means has led to the creation of large amounts of interaction data between learners and content, giving rise to tremendous potential for intelligent recommendation systems. Traditional collaborative filtering approaches are plagued by the lack of flexibility in expressing learners' learning preferences and also suffer from the cold-start problem. To address these deficiencies, this research presents a Multi-Layered Neural Collaborative Filtering (ML-NCF) algorithm for learner content recommendations that overcomes these shortcomings. In the presented ML-NCF framework, GMF is combined with a deep MLP architecture that learns both linear and nonlinear user-item interactions along with a self-attention mechanism for dynamically assigning weights to different contexts in learning. The effectiveness of the approach was evaluated using two widely used datasets: the MOOC Interaction dataset (N = 94,506 learner-content interactions) and the EdNet dataset (N = 131,441 interactions). ML-NCF resulted in 87.4% hit rate, NDCG at 0.723, precision at 0.814, recall of 0.791, and MRR of 0.768, showing superior performance compared to baseline models such as BPR-MF, NeuMF, and LightGCN by 9.2%, 6.8%, and 4.1% in hit rate, respectively. The effectiveness of all components of the proposed architecture has been validated through an ablation study. It is evident from the findings that the ML-NCF approach improves educational recommendations by leaps and bounds. This research provides strong support for the application of the ML-NCF system in adaptive learning systems, intelligent tutoring, and e-learning applications. The future direction will include federated learning mechanisms in recommendation scenarios.