Personal Recommendation Based on Weighted Bipartite Networks

Recently, network based recommendation algorithms have demonstrated much better performance than the standard collaborative filtering method, and most of which have been focused on the unweighted cases even in a multigraded rating system. However, these modifications from multigraded rating data to binary data may lose information, thus hinder the expressing of user's preference and finally misleading the recommendation systems. In this paper, we propose to use weighted bipartite user-object networks to model the recommender systems. The weight of the edge is directly the rate that a user giving on an object. We use a benchmark dataset, i.e., Moivelens dataset, to test the performance. The results show that weighted theme has higher recommendation accuracy than its unweighted counterpart.