A Fast Algorithm for Hunting State-Backed Twitter Trolls

Abstract

In recent years, state-backed troll accounts have been adopted extensively by many political parties, organizations, and governments to negatively influence political systems, persecute perceived opponents, and exacerbate divisiveness within societies. Thus, the need for an automatic state-backed troll classification system has increased. Various algorithms have been proposed in the literature to handle this problem, but a majority of them consider all types of trolls as one type which decreases the performance of classification algorithms. Our goal in this paper is to design a thorough method for detecting state-backed trolls on Twitter with the ability to work efficiently in any case regardless of the language, the location, and the purpose of the troll account. For accurate classification, a set of novel effective and powerful features from various categories are proposed. To train our algorithm, we gathered a large and relevant dataset from Twitter. The results show that the proposed algorithm achieves high classification accuracy (approximately 99%) and has the ability to classify state-backed troll accounts regardless of the language or the location of the account.