Towards Revealing Parallel Adversarial Attack on Politician Socialnet of Graph Structure

Socialnet becomes an important component in real life, drawing a lot of study issues security and safety. Recently, for the features graph structure socialnet, adversarial attacks on node classification are exposed, automatic attack methods such as fast gradient (FGA) NETTACK developed per-node attacks, which can be utilized multinode sequential way. However, due to overlook perturbation influence between different above method does not guarantee global success rate all target nodes, under fixed budget perturbation. In this paper, we propose parallel framework classification. We redesign new loss function objective nonconstraint constraint perturbations, respectively. Through constructing intersection supplement mechanisms then integrate filtering-based P-FGA P-NETTACK unified framework, finally realizing attacks. Experiments politician socialnet dataset Polblogs with detailed analysis conducted show effectiveness our approach.