An Efficient Victim Prediction for Sybil Detection in Online Social Network

With the rapid development of Online Social Networks (OSNs), OSNs have become a rewarding target for attackers. One particularly representative attack is the Sybil attack, Sybil accounts create a lot of malicious activities, which poses a serious threat to the safety of normal users. Many existing Syibl detection mechanisms have preconditions or assumptions, for example, limiting the number of attacking edges. But in general, the assumption is only a handful, often does not hold in real life scenarios. When the assumption is not established, these mechanisms perform poorly. In this work, We propose a scheme that uses victim prediction to improve Sybil detection accuracy. And our solution does not need to be based on any assumptions. First, we designed a victim classifier to predict victims. Then, based on the prediction results, the edge weights in the graph model are modified. Next, trust propagation is performed on the graph model. Finally, sorting all accounts. The experimental results show that our scheme can ensure that the majority of normal users rank higher than Sybils, thus classifying normal users and Sybils.