Parallel Community Detection on Massive Graphs

Community detection groups a network into node sets according to their connections. It is an effective way to understanding and analyzing graph-structured data, such as social networks, collaboration networks, and bioinformatic networks. With the flourishing development of social network applications, it has become more desirable to explore graphs from a community-level view. However, based on sequential algorithms, most existing community detection methods are not suitable for massive graphs. In this paper, we propose a Parallel Community Detection approach, named ParCoDe. Just like the native sequential algorithm, it uses "community modularity" as the metric. The detecting process starts from each single node and performs in a bottom-up way. In order to improve its performance, we propose an approximate solution to accelerate the speed of detection with little loss of accuracy. We have implemented ParCoDe on Giraph. Comprehensive experiments on both real and synthetic datasets demonstrate that ParCoDe is of well scalability and is efficient for community detection.