Article Preview
TopDue to the increasing importance of online social networks, the misuse of online service accounts by attackers has been a rising problem. As a result, there has been numerous studies on the topic by researchers.
Many researchers focus on attacks on the network such as Sybil (Boshmaf et al., 2015; Al-Qurishi et al., 2017). Sybil attack is where attackers create a large number of fake accounts, and use them to manipulate a network or community. These fake accounts can be used for various nefarious purposes including distributing spam, distributing malware, launching phishing campaigns, spreading misinformation, and conducting opinion frauds. To counter this, many researchers focused on the detection of fake accounts.
Boshmaf et al. (2015) proposed a framework known as Integro that takes a graph-based approach to this problem. The social network is modeled as a graph where nodes represent users and edges represent connection/friendship between users. The nodes and the edges are assigned weights based on user profile information and graph structure. They observed a certain patterns from the social graphs, such as the fake accounts’ tendency to befriend as many real accounts as possible from disparate communities, and the existence of certain vulnerable users/nodes that are more likely to befriend fake accounts. Their proposed system abstracts the problem of detecting fake accounts as a trust propagation problem (Boshmaf et al., 2015).
Al-Qurishi et al. (2017) proposed a system called SybilTrap for detecting fake accounts that takes a graph-based semi-supervised learning approach. First, social interaction graphs were created where nodes again represent users and edges represent user interactions (e.g., direct mentions, replies, or retweets). As an improvement to Integro, SybilTrap considers post content in addition to profile information and graph structure for assigning node and edge weights. Finally, a label propagation algorithm is performed and nodes are ranked based on how likely they are to be fake accounts (Al-Qurishi et al., 2017).