Alleviating the data sparsity problem of recommender systems by clustering nodes in bipartite networks

Abstract

Recommender systems help users to find information that fits their preferences in an overloaded search space. Collaborative filtering systems suffer from increasingly severe data sparsity problem because more and more products are sold in commercial websites, which largely constrains the performance of recommendation algorithms. User clustering has already been applied to recommendation on sparse data in the literature, but in a completely different way. In most existing works, user clustering is directly used to identify the similar users of the target user to whom we want to make recommendation. More specifically, the users who are clustered in the same group of the target user are considered as similar users. However, in this paper we use user clustering to reconstruct the user-item bipartite network such that the network density is significantly improved. The recommendation made on this dense network thus can achieve much higher accuracy than on the original sparse network. The experimental results on three benchmark data sets demonstrate that, when facing the problem of data sparsity, our proposed recommendation algorithm based on node clustering achieves a significant improvement in accuracy and coverage of recommendation.