UCLA

Community clustering is well-studied in the context of social network analysis. However, in web services such as online retailing, music sharing library and social news website, the user data are more naturally modeled by the bipartite networks, where users are one class of the nodes and the product, song or story are the other. In this work, we propose a logistic weight model to transform the bipartite network to a weighted uni-partite network for clustering purpose. We experiment the model on Balatarin.com, a social news website and show the model leads to remarkably better clustering results by identifying and exploiting more informative users from the data.