"Breaking Bad": Ranking Bot’s Malevolence in Twitter

Typically, bots are faster in exploiting social media than human users. While some bots are benign in nature, others are malicious and invasive; they show suspicious behavior and become involved in promotional activities that eventually lead to their suspension. Therefore, there is a need to shift the focus from automated bots detection to malicious bots detection since not all bots are bad; many of the bots are created for entertainment and customer services purposes (e.g. chat-bot).In this paper, we propose a real-time ranking system to rank bots based on the degree of their malevolence. We characterize bot behavior with different features measuring their media bias, hate content, and suspicious URLs. We use a deep learned model to produce a ranking score. The model is trained on a novel dataset of suspended bots tracked over four months on Twitter. Our model achieves 94.29% ranking-based average precision in ranking the bots and can predict malicious bots successfully.