Dynamic reputation information propagation based malicious account detection in OSNs

People all around the world have become increasingly dependent on online social networks (OSNs), meanwhile, the number of malicious accounts in OSNs is also rapidly growing. Traditional content-based data mining techniques and user graph-based methods are asking for more and more computing resources from networks providers, especially for the networks with huge and complicated network topologies. Moreover, traditional content-based analysis methods need to keep up with the times, which need to be retrained when the structure of users’ data changes or when the malicious contents come along with some pop cultures. With the purpose of reducing the dependence on network providers’ computing resources and improving the precision of detection, a novel detection method of malicious account, which bases on dynamic users’ reputation information propagation, is proposed in this paper. According to the comparison result of requesting user’s comprehensive reputation and malicious threshold, user can mark requesting user’s reputation so as to achieve the purpose of malicious account detection and providing indirect recommended reputation information about requesting user for other users. Through experiments with two real-world datasets and comparison with two typical efficient detection algorithms, this algorithm can effectively detect malicious accounts without the central detection system as well as improve the detection precision.