A Hybrid Recommender System for Pedagogical Resources

Abstract

AbstractThe present paper proposes an e-learning system that combines popularity and collaborative filtering techniques to recommend pedagogical resources. A recommender system helps users get a correct and personalized decision by applying several recommendation methods such as content-based, collaborative filtering, and other hybrid approaches. However, predicting a relevant resource with a specific context, like pedagogical content, becomes a challenge. In our work, we propose a model to ameliorate the traditional collaborative filtering technique by (i) using the Singular Value Decomposition (SVD) to tackle the problem of scalability and data sparsity; (ii) extracting the most popular resources that the user does not interact with before to resolve the cold start problem; and (iii) combining the results of popularity and SVD factorization methods to improve the recommendation accuracy that evaluated by applying the recall, precision and f1-score of each approach. The comparison shows that the obtained results exhibit an encouraging performance of our model.KeywordsRecommender systemE-learningCollaborative filteringPopularitySVD