Community Detection in Large Directed Graphs.

Many real-life systems are represented by directed networks but most of the existing community detection research is focused on undirected networks. We present here a new scalable Map-Reduce algorithm to discover PageRank-driven hierarchical community structures in directed graphs. Main features of our approach are that we seek to find a directed community centered around any selected core node of a directed graph, and a community is defined by two hierarchical structures connected to the core node. An upper structure, that points "to" the core node, and a lower structure that is pointed to "by" the core node. Nodes chosen for inclusion in our hierarchical community structure must display importance ratings (PageRank) very similar to the rating of the selected core node. We have demonstrated here that our algorithm finds meaningful hierarchical communities in real-world data sets. We have also evaluated the effects of the PageRank of the core node and the PageRank threshold parameter for the nodes included in its communities.