AN INVESTIGATION OF BOT ACCOUNT IDENTIFICATION ON TWITTER DATA

Abstract

Twitter bots are automated accounts which are programmed to perform certain tasks which resemble those of daily active users such as tweeting, liking a tweet or following. Most of the time, they are designed with malicious intent such as spreading fake news, spamming or manipulation of public opinion. More and more, the Twitter platform is being constantly threatened by malicious bots. The aim of this specific research is to investigate different state-of-the-art systems and corresponding attributes to identify bots on Twitter data and to find how identification can be improved. Evaluations are performed on real world data using both learning and non-learning systems demonstrating the performance of bot identification using different attributes, classifiers, and publicly available tools. Results are very promising when the aforementioned systems are trained and tested on data with the ground truth compared to the existing literature with/without learning systems.