Bot Detection on Online Social Networks Using Deep Forest

Nowadays, social networks are widely used not only by humans, but also by bots (automated agents). Recent studies have reported that bot accounts are used across different dimensions and in different granularities (e.g. performing terrorist propaganda, spreading misinformation and polluting content). Bot detection has become a significant challenge, especially on online social networks. Today, researchers through Twitter are attempting to propose approaches for bot detection. However, they are confronted with certain challenges owing to the problems inherent to text and the use of language-dependent features. Therefore, we defined a set of statistical features which proved their importance in our work. Our proposed features are based on how much the others interact with the posts of the user, when the user interacts and how much the user interacts. We demonstrated that the mere use of information from the metadata of user profiles and the metadata of posts with a Deep Forest algorithm is sufficient in order to detect bot accounts accurately. In fact, this yielded an Accuracy result of 97.55%.