Systematic Literature Review: Comparison on Collaborative Filtering Algorithms for Recommendation Systems

Abstract

The recommendation system is divided into collaborative filtering (CF), content-based (CB), and hybrid approaches. This paper focuses on the CF approach which has many algorithms such as K-Nearest Neighbor (KNN), K-Means, Singular Value Decomposition (SVD), etc. We used the systematic literature review approach to gather papers related to CF and 28 research papers were eventually considered for analysis in KNN, deep learning, and SVD. From the review results, most of the datasets used in CF were movie datasets to test the recommendation model, and most of the models produced a good result in recommending items. To achieve good results, the majority of existing works combine more than one method to overcome or reduce the impact of CF problems (cold-start, sparsity, shilling attacks, etc.) which can affect the recommendation performance.