Title:A Distributed Algorithm for Overlapped Community Detection in Large-Scale Networks

Abstract:Overlapped community detection in social networks has become an important research area with the increasing popularity and complexity of the networks. Most of the existing solutions are either centralized or parallel algorithms, which are computationally intensive - require complete knowledge of the entire networks. But it isn't easy to collect entire network data because the size of the actual networks may be prohibitively large. This may be a result of either privacy concerns or technological impediments. Performing in-network computation solves both problems utilizing the computational capability of the individual nodes of the network. Simultaneously, nodes communicate and share data with their neighbors via message passing, which may go a long way toward mitigating individual nodes' privacy concerns in the network. All the aforementioned concerns motivated us to design a decentralized or distributed technique to detect overlapped communities in a large-scale network. It is desirable because this technique does not offer a single point of failure, and the system as a whole can continue to function even when many of the nodes fail. In this paper, we address the overlapped community detection problem for large-scale networks. We present an efficient distributed algorithm, named DOCD, to identify the overlapped communities in the network. The DOCD algorithm's efficiency is verified with extensive simulation study on real network data such as Dolphin, Zachary karate club, Football club, and Facebook ego networks. We show that DOCD algorithm is capable of keeping the asymptotically same results with the existing classical centralized algorithms in terms of community modularity and the number of identified communities. The DOCD algorithm can also efficiently identify the overlapped nodes and overlapped communities with a small number of rounds of communication and computation.