Propagation mergence for community detection

Abstract

The Louvain method is a successful technique for community detection that decomposes a network by optimizing the modularity of the partitions. It was not designed, however, for cases where a network is incomplete and network nodes contain some unused information. A new method is introduced here, Propagation Mergence (PM), that was designed to handle incomplete networks, and leverage node information to improve community detection. PM utilizes PageRank, propagation, and greedy merging to 1) locate local hubs, 2) discover small subgraphs, and then 3) create larger subgraphs via merging. An illustrative literature network example is presented to demonstrate that PM yields results that improve upon the Louvain method.