Combating Fraud in Online Social Networks: Characterizing and Detecting Facebook Like Farms

As businesses increasingly rely on social networking sites to engage with their customers, it is crucial to understand and counter reputation manipulation activities, including fraudulently boosting the number of Facebook page likes using so-called like farms. Thus, social network operators have started to deploy various fraud detection algorithms such as graph clustering methods, however, with limited efficacy. In fact, this paper presents a comprehensive analysis and evaluation of existing graph-based fraud detection algorithms for detecting like farm accounts. Our results show that more sophisticated and stealthy farms can successfully evade detection by spreading likes over longer timespans and by liking many popular pages to mimic normal users. Next, we analyze a wide range of features extracted from users' timeline posts, which we group into two main classes: lexical and interaction-based. We find that like farm accounts tend to more often re-share content, use fewer words and poorer vocabulary, target fewer topics, and generate more (often duplicate) comments and likes compared to normal users. Using these timeline-based features, we experiment with machine learning algorithms to detect like farms accounts, obtaining appreciably high accuracy (as high as 99% precision and 97% recall).