Movies recommendation networks as bipartite graphs

Abstract

In this paper we investigate the users' recommendation networks based on the large data set from the Internet Movie Database. We study networks based on two types of inputs: first (monopartite) generated directly from the recommendation lists on the website, and second (bipartite) generated through the users' habits. Using a threshold number of votes per movie to filter the data, we actually introduce a control parameter, and then by tuning this parameter we study its effect on the network structure. From the detailed analysis of both networks we find that certain robust topological features occur independently from the value of the control parameter. We also present a comparison of the network clustering and shortest paths on the graphs with a randomized network model based on the same data. © 2008 Springer-Verlag Berlin Heidelberg.