SybilFlyover: Heterogeneous graph-based fake account detection model on social networks

Abstract

Organized social robot accounts can launch Sybil attacks on online social networks (OSNs) for various malicious purposes, thus significantly affecting the user experience of online communities and damaging the reputation of OSNs. Therefore, detecting Sybil accounts in OSNs is crucial in cyberspace governance. Among the existing Sybil account detection methods, structure-based methods appear to be the most promising because of their ability to model the states and behaviors of social network accounts effectively. However, most of these structure-based methods only model the social accounts in OSNs while ignoring the content generated in OSNs, such as tweets of Twitter accounts. To address this deficiency, this study proposes an efficient model called SybilFlyover, which uses a heterogeneous graph-based method to represent all types of entities existing in an OSN uniformly, as well as complex relationships between the entities in a directed heterogeneous social network graph. During the modeling process, content-based social network information is injected into the model using a method based on prompt learning to achieve more accurate modeling of the real state of an OSN. Finally, the social network graph was processed using a transformer-based method to identify nodes representing Sybil accounts. The results of an experiment on a real public dataset demonstrate that the proposed SybilFlyover model outperforms existing state-of-the-art baseline models in Sybil account detection.