In the past decade, network structures have penetrated nearly every aspect ofour lives. The detection of anomalous vertices in these networks has becomeincreasingly important, such as in exposing computer network intruders oridentifying fake online reviews. In this study, we present a novel unsupervisedtwo-layered meta-classifier that can detect irregular vertices in complexnetworks solely by using features extracted from the network topology.Following the reasoning that a vertex with many improbable links has a higherlikelihood of being anomalous,we employed our method on 10 networks of variousscales, from a network of several dozen students to online social networks withmillions of users. In every scenario, we were able to identify anomalousvertices with lower false positive rates and higher AUCs compared to otherprevalent methods. Moreover, we demonstrated that the presented algorithm isefficient both in revealing fake users and in disclosing the most influentialpeople in social networks.
展开▼
