Abstract
Sybil attacks are increasingly prevalent in online social networks. A malicious user can generate a huge number of fake accounts to produce spam, impersonate other users, commit fraud, and reach many legitimate users. For security reasons, such fake accounts have to be detected and deactivated immediately. Various defense schemes have been proposed to deal with fake accounts. However, most identify fake accounts using only the structure of social graphs, leading to poor performance. In this paper, we propose a new and scalable defense scheme, SybilTrap. SybilTrap uses a semi-supervised technique that automatically integrates the underlying features of user activities with the social structure into one system. Unlike other machine learning-based approaches, the proposed defense scheme works on unlabeled data, and it is effective in detecting targeted attacks, because it manipulates different levels of features of user profiles. We evaluate SybilTrap on a dataset collected from Twitter. We show that our proposed scheme is able to accurately detect Sybil nodes as well as huge conspiracies among them.