Research in the International Journal of Computational Systems Engineering introduces a hybrid recommendation model that could help with one of the common challenges facing universities offering online courses. How to recommend the most appropriate course for prospective students.
The approach uses Naive Bayes classification and collaborative filtering to improve accuracy and personalised course suggestions. This, the researchers suggest, could ultimately enhance the learning experience for students.
Online course recommendation systems have long struggled with issues such as the “cold start” problem, data sparsity, and inadequate personalisation. The “cold start” problem occurs when a recommendation system lacks sufficient historical data about new users or courses, making it difficult to provide relevant suggestions. Data sparsity, on the other hand, refers to the limited amount of data available for each course, which can hinder the system’s ability to capture students’ preferences. Additionally, inadequate personalisation leads to generalised recommendations that may not match the unique needs of individual students, resulting in a less effective user experience.
The hybrid model discussed in IJCSE could resolve these issues. By using Naive Bayes classification, it can predict the likelihood that a particular course aligns with the interests of a given student based on course features. Collaborative filtering then examines patterns in student character and identifies similar users to recommend courses based on what others with similar learning habits have chosen.
The system also adds a dynamic weight adjustment feature that adjusts the model’s recommendations depending on whether a student is a new user or an experienced one. This mechanism improves the precision and diversity of the suggestions, ensuring that the system remains useful for all types of students.
The team tested the system with data from 25,000 students and 1,000 courses. Compared to traditional methods, it demonstrated a 12% improvement in Precision@10 (the percentage of relevant courses within the top 10 recommendations) and a 10.5% improvement in Recall@10 (the percentage of relevant courses among the top 10 recommendations). Most notably, in cold start scenarios, the hybrid model significantly outperformed deep neural networks. Even with a data sparsity of 98%, the hybrid model’s accuracy fell at half the rate of traditional algorithms.
Chen, Z. and He, M. (2026) ‘Research on integrating naive Bayes and collaborative filtering into an online-course recommendation model for universities’, Int. J. Computational Systems Engineering, Vol. 10, No. 6, pp.12–21.
No comments:
Post a Comment