Article: Research on integrating naive Bayes and collaborative filtering into an online-course recommendation model for universities Journal: International Journal of Computational Systems Engineering (IJCSYSE) 2026 Vol.10 No.6 pp.12 - 21 Abstract: Current college online course recommendation systems struggle with cold start, data sparsity, and limited personalisation, reducing recommendation accuracy and user satisfaction. This study proposes a hybrid model combining naive Bayes and collaborative filtering to address these challenges. By integrating course metadata and user behaviour data, the model extracts multi-dimensional features, capturing both static preferences and dynamic behaviours through probabilistic modelling and collaborative filtering. Experiments on data from 25,000 students and 1,000 courses show that the model improves Precision@10 and Recall@10 by 12% and 10.5% respectively, compared to individual models. In cold-start scenarios, it achieves an F1@10 score of 0.35, compared to 0.27 for DNN. Under 98% sparsity, its accuracy degrades only half as much as traditional collaborative filtering. With 2.3 seconds per iteration and a 26.4% increase in click-through rate, the model demonstrates efficiency and effectiveness in personalised online course recommendations. Inderscience Publishers - linking academia, business and industry through research

Title: Research on integrating naive Bayes and collaborative filtering into an online-course recommendation model for universities

Authors: Zhongyao Chen; Min He

Addresses: Academic Affairs Office, Chongqing City Vocational College, Yongchuan District, Chongqing City, 402160, China ' College of Information and Intelligent Manufacturing, Chongqing City Vocational College, Yongchuan District, Chongqing City, 402160, China

Abstract: Current college online course recommendation systems struggle with cold start, data sparsity, and limited personalisation, reducing recommendation accuracy and user satisfaction. This study proposes a hybrid model combining naive Bayes and collaborative filtering to address these challenges. By integrating course metadata and user behaviour data, the model extracts multi-dimensional features, capturing both static preferences and dynamic behaviours through probabilistic modelling and collaborative filtering. Experiments on data from 25,000 students and 1,000 courses show that the model improves Precision@10 and Recall@10 by 12% and 10.5% respectively, compared to individual models. In cold-start scenarios, it achieves an F1@10 score of 0.35, compared to 0.27 for DNN. Under 98% sparsity, its accuracy degrades only half as much as traditional collaborative filtering. With 2.3 seconds per iteration and a 26.4% increase in click-through rate, the model demonstrates efficiency and effectiveness in personalised online course recommendations.

Keywords: naive Bayes; collaborative filtering; online course recommendation; cold start; data sparsity.

DOI: 10.1504/IJCSYSE.2026.152654

International Journal of Computational Systems Engineering, 2026 Vol.10 No.6, pp.12 - 21

Received: 10 Jun 2025
Accepted: 22 Jan 2026
Published online: 01 Apr 2026 *

Title: Research on integrating naive Bayes and collaborative filtering into an online-course recommendation model for universities

Keep up-to-date