Disintegrating constant communities in complex networks

Abstract

Community detection in complex networks has been one of the major research areas since last one decade. Most of the community detection algorithms are non-deterministic, thus yielding different community structures in different iterations of an algorithm for a certain network. However, recent research has showed the existence of Constant Communities (CCs) — the core of a community structure, whose members always remain consistent irrespective of any iteration of the algorithm. Therefore, from an adversarial point of view, disintegrating such CCs requires thorough understanding of the internal connectivity among nodes as it is even harder to break a CC compared to breaking a community. Here we focus on how small disruption of the structure of a CC can cause major disintegration of its structure. Specifically, we propose a novel ranking scheme of edges forming a CC based on several factors: size and cliquishness of the CC as well as number of fragments and evenness of splits caused by the removal of an edge. We use several (and diverse) networks and compare the proposed ranking scheme with state-of-the-art edge-based ranking schemes to show that our ranking scheme is different from others and highly effective. Moreover, to understand the physical significance of edges forming CCs, we interpret our results on real-world networks and observe that edges that are ranked higher by our scheme are highly vulnerable to the attack that can cause maximum disintegration of the CCs.