Content-Based Personalized Course Recommendation in e-Learning Ecosystems: A TF-IDF and Similarity Measures Approach

Abstract

The rapid expansion of online education platforms has posed significant challenges for learners in identifying courses aligned with their goals and interests. This paper proposes a novel content-based Course Recommender System (CRS) tailored for e-learning ecosystems, specifically addressing cold-start scenarios without user history. The innovation lies in integrating Term Frequency–Inverse Document Frequency (TF-IDF) and Count Vectorization with Cosine and Jaccard Similarity measures to create a balanced framework that optimizes accuracy, recall, and diversity, with Jaccard enhancing exploratory recommendations as validated by statistical analysis. Evaluated on a dataset of 3682 Udemy courses across diverse subjects (Business, Graphic Design, Musical Instruments, Web Development), the system’s performance was assessed using Precision@k, Recall@k, Mean Average Precision (MAP), Mean Reciprocal Rank (MRR), diversity index, and accuracy metrics. Results show the TF-IDF with cosine similarity model achieving 99.98% accuracy at top-10 recommendations, while Jaccard-based models enhance diversity (diversity index score of 0.85), confirming the approach’s scalability and robustness. These findings contribute to personalized course discovery in large-scale e-learning environments.