One Improved Collaborative Filtering Algorithm Based on Bipartite Network Structure

Abstract

In this paper, we propose a novel method based on classical Collaborative Filtering (CF) and biparitite network, structure. Different from the CF method, item similarity is viewed, as item recommendation power or the item popularity for the item in this system. Then, we redistribute the item similarity equallily, to other items as their inital resource by taking bipartite network structure into account. In our benchmark dataset, our method demonstrates us with a good performance in rank value, improving 12\% than the CF method. Furthermore, a free parameter $\beta$ is introduced to tune the contribution of the item similairty to keep our method more scalable. Numerical results demonstrates that the algorithm performance can improved on different measurements with diffent $\beta$.