Similarity-Based and Sybil Attack Defended Community Detection for Social Networks

Abstract

Community detection is a fundamental problem in analyzing the network building principles. Previous community detections left many security issues that every user in a social network must be assigned into a community, which enables attackers easily infiltrate into a community at little cost. For example, an online cancer patients’ community consists solely of cancer patients that want to share drug and treatment information together, but many malicious sybil attackers who collude in joining a target community with ulterior purpose will be probably classified into the target community if one or more patients randomly accept their friendship requests from these attackers. This may lead to the privacy disclosure of community information. Thus, to address this problem, this brief focuses on the novel mechanism of sybil-attack-defended community detection. Firstly, we propose a new similarity-based community detection algorithm (SCDA) which can incorporate any similarity metric to efficiently detect communities and defend sybil attacks in multi-level granularities. Secondly, we are the first to consider both the hierarchical community detection and sybil-attack defense. Several possible sybil-attack models are defined to simulate the attack process in real social networks. Finally, we conduct extensive experiments to demonstrate that our method has a good performance both on community detection and sybil-attack defense.