Integrating Clustering and Ranking on Hybrid Heterogeneous Information Network

Recently, ranking-based clustering on heterogeneous information network has emerged, which shows its advantages on the mutual promotion of clustering and ranking. However, these algorithms are restricted to information network only containing heterogeneous relations. In many applications, networked data are more complex and they can be represented as a hybrid network which simultaneously includes heterogeneous and homogeneous relations. It is more promising to promote clustering and ranking performance by combining the heterogeneous and homogeneous relations. This paper studied the rankingbased clustering on this kind of hybrid network and proposed the ComClus algorithm. ComClus applies star schema with self loop to organize the hybrid network and uses a probability model to represent the generative probability of objects. Experiments show that ComClus can achieve more accurate clustering results and do more reasonable ranking with quick and steady convergence.