Weighted bipartite network and personalized recommendation

In this paper, the degree distributions of a bipartite network, namely Movielens, are investigated. The statistical analysis shows that the distribution of the degree product, ku ko, has an exponential from, where ku and ko denote the user and object degrees respectively. By introducing the edge weight effect on the recommendation performance, an improved recommendation algorithm based on mass diffusion (MD) process is presented. We argue that the edges weight of the user-object bipartite network should be taken into account to measure the object similarity. By taking into account the user and object degree correlations, the weighted bipartite network is constructed. The numerical results of the MD algorithms on the weighted network indicate that both of the accuracy and diversity could be increased at the optimal case. More importantly, we find that, at the optimal case, the edge weight distribution would change from the exponential form to the poisson form. This work may shed some light on how to improve the recommendation algorithm performance by considering the statistical properties.