Unsupervised detection of coordinated fake followers on social media

Social media accounts are known to have automated accounts, referred to as socialbots, among their active users. While these accounts are not necessarily harmful,they are widely used to manipulate engagement metrics and in coordinated disinformationcampaigns. In this work, we propose a novel unsupervised approach todetect a subset of social bots, known as fake followers, which are used to deceitfullyamplify online popularity of users. Our method identifies fake followers bydetecting anomalous temporal following patterns within the followers of a socialmedia account. Furthermore, we use our method to investigate the prevalence ofanomalous followers in the Turkish political Twitter network (later rebranded as X).In addition to detection, we also demonstrated that groups of anomalous followersmay act in coordination across several accounts in the same network. Our resultsshow that the proposed framework can be used to investigate large-scale coordinatedmanipulation campaigns on social media platforms.