Community Detection Boosts Network Dismantling on Real-World Networks

Abstract

Network dismantling techniques have gained increasing interest during the last years caused by the need for protecting and strengthening critical infrastructure systems in our society. We show that communities play a critical role in dismantling, given their inherent property of separating a network into strongly and weakly connected parts. The process of community-based dismantling depends on several design factors, including the choice of community detection method, community cut strategy, and inter-community node selection. We formalize the problem of community attacks to networks, identify critical design decisions for such methods, and perform a comprehensive empirical evaluation with respect to effectiveness and efficiency criteria on a set of more than 40 community-based network dismantling methods. We compare our results to state-of-the-art network dismantling, including collective influence, articulation points, as well as network decycling. We show that community-based network dismantling significantly outperforms existing techniques in terms of solution quality and computation time in the vast majority of real-world networks, while existing techniques mainly excel on model networks (ER, BA) mostly. We additionally show that the scalability of community-based dismantling opens new doors towards the efficient analysis of large real-world networks.