Performance Enhancement of Collaborative Filtering through the Fusion Algorithm of Singular Value Decomposition and Association Rule

Abstract

Collaborative filtering (CF) is the leading algorithm as a recommendation system. However, the algorithm falls short due to the limitations of sparsity and scalability. Sparsity problem comes when the user has few items purchased or reviewed. Scalability problem occurs under too many users and items. In this study, the limitations of CF are enhanced through the usage of singular value decomposition (SVD) and association rule (AR). AR searches for relevance among items, and SVD decreases the dimensionality of the data to be applied on CF. This hybrid CF algorithm, which is combined with SVD and AR, is compared with other recommendation algorithms. Items which are included in computation were chosen through AR. SVD solved the short falls of data sparsity of CF. Both item-based CF and user-based CF are considered. The hybrid algorithm showed better performance than other recommendation systems under sparsity and scalability problem.