Community Detection Attack in Complex Networks: A Multiobjective Perspective

The abuse of community detection algorithms posed significant risks of privacy leakage. To protect personal privacy in complex networks, community detection attack (CDA) algorithms have been proposed, which modify a small number of connections to obscure the original community structure. However, most of the existing studies focus on designing effective attack strategies while the attack budget should be given by decision makers in advance. In this article, we transform CDA into a biobjective optimization problem, where the attack effectiveness and the attack budget are optimized simultaneously. To solve this problem, an effective multiobjective evolutionary algorithm (MOEA) named CDA-MOEA is proposed, which provides the decision maker with a holistic view for assessing attack strategies. Additionally, a budget-aware population repair operator based on betweenness and permanence is suggested to enhance the diversity and quality of the nondominated solutions in CDA-MOEA. Experimental evaluations are conducted by comparing CDA-MOEA with four state-of-the-art baseline methods against five community detection algorithms on eight real-world networks. The results indicate that CDA-MOEA significantly improves attack effectiveness at the same attack cost comparing to baseline methods, confirming its superiority and practical applicability.