A one-class classification approach for bot detection on Twitter

Abstract Twitter is a popular online social network with hundreds of millions of users, where n important part of the accounts in this social network are not humans. Approximately 48 million Twitter accounts are managed by automated programs called bots, which represents up to 15% of all accounts. Some bots have good purposes, such as automatically posting information about news and academic papers, and even to provide help during emergencies. Nevertheless, Twitter bots have also been used for malicious purposes, such as distributing malware or influencing the perception of the public about a topic. There are existing mechanisms that allow detecting bots on Twitter automatically; however, these mechanisms rely on examples of existing bots to discern them from legitimate accounts. As the bot landscape changes, with the bot creators using more sophisticated methods to avoid detection, new mechanisms for discerning between legitimate and bot accounts are needed. In this paper, we propose to use one-class classification to enhance Twitter bot detection, as this allows detecting novel bot accounts, and requires only from examples of legitimate accounts. Our experiment results show that our proposal can consistently detect different types of bots with a performance above 0.89 measured using AUC, without requiring previous information about them.