A Matrix Factorization Method for Clustering in Heterogeneous Information Networks

Non-negative matrix factorization (NMF) has become quite popular recently on the relational data due to its several nice properties and connection to probabilistic latent semantic analysis (PLSA). However, few algorithms take this route for the heterogeneous networks. In this paper we propose a novel clustering method for heterogeneous information networks by searching for a factorization that gives compatible clustering solutions across multiple sub-networks. Moreover, we develop an automatic weight learning strategy in order to balance the effects of different sub-networks brought to the consensus. Experimental results on realworld dataset demonstrate the effectiveness of our approach.