Development of a Hybrid-Model Recommender System for an E-Learning Platform

Abstract

E-learning systems continue to experience a growing number of courses and learning materials, making it increasingly difficult for learners to locate resources that are relevant to their interests, capabilities, and learning goals. This study developed and evaluated a lightweight hybrid recommender system for an e-learning platform that integrates content-based filtering (CBF) and collaborative filtering (CF) to improve the relevance and personalization of learning-resource recommendations. The content-based component employed Term Frequency-Inverse Document Frequency (TF-IDF) and cosine similarity to identify resources based on their content, while the collaborative filtering component used Singular Value Decomposition (SVD) to model learner-item interaction patterns. Experimental evaluation was conducted using accuracy, precision, recall, F1-score, and processing time. The hybrid model achieved the best overall performance, recording an accuracy of 0.88, precision of 0.89, recall of 0.80, and F1-score of 0.84, compared with the content-based model, which achieved 0.74, 0.62, 0.57, and 0.60, respectively, and the collaborative filtering model, which achieved 0.81, 0.74, 0.67, and 0.70, respectively. Although the hybrid model recorded a processing time of 0.068 seconds, slightly higher than collaborative filtering (0.064 seconds) and comparable to content-based filtering (0.067 seconds), the improvement in recommendation quality outweighed the marginal increase in execution time. The results demonstrate that integrating content-based and collaborative filtering can reduce the effects of cold-start and data sparsity while producing more relevant and consistent recommendations. The proposed hybrid model therefore provides an effective and computationally lightweight approach for personalized e-learning recommendation in resource-constrained educational environments.