Dynamic peer learning recommendations in e-learning using hybrid collaborative filtering and interaction-based clustering

Introduction Peer-learning recommendation remains an open challenge in e-learning systems, as most existing approaches—such as matrix factorization and neural collaborative filtering—rely on static interaction patterns. These methods often ignore contextual information including learner roles, content difficulty, and temporal engagement behavior. As a result, they struggle to form meaningful peer groups or provide adaptive learning paths that align with pedagogical needs. Methods To address these limitations, we propose a hybrid context-aware peer learning recommender that integrates collaborative filtering with interaction-based clustering. The framework incorporates adaptive peer group formation using multiple loss functions and multifactor BERT embeddings to capture content semantics. In addition, learner-specific characteristics such as difficulty level, job role, and software skills are explicitly modeled. These contextual and semantic features are dynamically used to cluster learners and generate personalized peer recommendations. Results and discussion Experiments conducted on an e-learning dataset demonstrate that the proposed model significantly outperforms sequential baseline approaches, as well as traditional matrix factorization and neural collaborative filtering models. The hybrid approach achieves an accuracy of 0.80, precision of 0.80, recall of 0.06, and an F1-score of 0.11. These results indicate improved personalization and contextual relevance in peer recommendations, enabling more adaptive and pedagogically suitable peer learning experiences.