Community Detection in Node Attributed Networks: A Late-fusion Approach

Abstract

With the burgeoning of online social media and the deluge of information in today's "big data" era, traditional community mining that relies on the connections of the nodes no longer suffices to find communities where the attributes of these nodes play an important role. Though vast research has been done to incorporate attribute information in search of network communities, few have focused on the late-fusion approach, where two partitions of a network are identified with traditional community detection and clustering algorithms respectively and are later combined to produce the final communities. We propose a new late-fusion method that assimilates two sources of information by creating an integrated graph whose edges represent the agreement of communities coming from the two partitions. We design a new technique to cope with networks with binary or categorical attributes in a way that clusters reflecting node similarities are found by a community detection algorithm on a virtual graph. We introduce a weighting parameter to allow for leveraging the strength between node connections and attributes. We experimentally demonstrate the performance of our method on various synthetic and real networks. We show that our late-fusion method comes as a flexible, accurate and efficient solution to the problem of community detection in attributed networks.