Collaborative Filtering Based on Non-Negative Matrix Factorization for Programming Problem Recommendation

Muepu Mukendi Daniel, Yutaka Watanobe, Md. Mostafizer Rahman

Abstract: This paper explores the use of Non-Negative Factorization for adapting collaborative filtering to programming exercises. Traditional collaborative filtering uses user preferences as ratings, but user feedback is expressed differently in programming exercises. The proposed approach captures user satisfaction based on the number of attempts and the final verdict. The NMF algorithm is used to factorize a non-negative input matrix into two non-negative matrices, and user-based collaborative filtering is implemented to make recommendations. The proposed model achieved an Mean Absolute Error of 0.06, Root Mean Squared Error of 0.2, a Recall of 51%, and a Precision of 54%, demonstrating its capability to make accurate and relevant recommendations for programming exercises.