Local Community Extraction for Non-overlapping and Overlapping Community Detection

Abstract

The scale of current networked system is becoming increasingly large, which exerts significant challenges to acquire the knowledge of the entire graph structure, and most global community detection methods often suffer from the computational inefficiency. Local community detection aims at finding a community structure starting from a seed vertex without global information. In this article, we propose a Local Community Extraction algorithm (LCE) to find the local community from a seed vertex. First, a local search model is carefully designed to determine candidate vertices to be preserved or discarded, which only relies on the local/incomplete knowledge rather than the global view of the network. Second, we expand LCE for the global non-overlapping community detection, in which the labels of detected local communities are seen as vertices’ attributive tags. Finally, we adopt the results of LCE to calculate a membership matrix, which can been used to detect the global overlapping community of a graph. Experimental results on four real-life networks demonstrate the advantage of LCE over the existing degree-based and similarity-based local community detection methods by either effectiveness or efficiency validity.