Dynamic and decentralized learning of overlapping network communities

Network communities exist as clusters of nodes whose intra-edge connectivity is stronger than edge connectivities between nodes from different clusters. Among others, identification of hidden communities unveils shared functional roles in biological networks, and assigns individuals in social networks to consumer groups for more targeted advertising. This is a rather challenging task in large-size networks due to temporal evolution of the underlying topology, presence of distortive anomalies, as well as the sheer scale of network data, often stored in distributed clusters. Most contemporary approaches resort to batch centralized processing, and are ill-equipped to address the afore-mentioned challenges. The present paper develops a novel decentralized algorithm for tracking overlapping communities in dynamic networks, while compensating for distortions due to anomalous nodes. Experiments conducted on global trade flow data demonstrate the efficacy of the proposed approach.