The use of restricted Boltzmann machines for clustering collaborative filtering

Abstract

<div><p class="Abstract">Collaborative filtering-based recommender systems have been successfully serviced through commercial online systems to assist people with searching the information useful to them. However, several problems inherent in such systems still exist, although a lot of research work has been devoted to finding solutions. This work focuses on clustering collaborative filtering to address the scalability problem. It proposes a novel method to determine the clustering criteria for enhancing the prediction and recommendation accuracy of the systems, which is typically degraded when the clustering algorithm is integrated into collaborative filtering. We use a restricted Boltzmann machine to find the genre preference of users, which is then inputted into the clustering algorithm to cluster users. Various experiments are conducted to evaluate the performance of the proposed method. As a result, our method showed superior performance in terms of various performance criteria compared to previous clustering collaborative filtering methods and some of the major traditional systems.</p></div>