DNA-inspired characterization and detection of novel social Twitter spambots

Spambot detection is a must for the protection of cyberspace, in terms of both threats to sensitive information of users and trolls that may want to cheat and influence the public opinion. Unfortunately, new waves of malicious accounts are characterized by advanced features, making their detection extremely challenging. In contrast with the supervised spambot detectors largely used in recent years and inspired by biological DNA, we propose an alternative, unsupervised detection approach. Its novelty is based on the idea of modeling online user behaviors with strings of characters representing the sequence of the user’s online actions. Exploiting this nature-inspired behavioral model, the proposed technique lets groups of spambots emerge from the crowd, by comparing the accounts’ behaviors. Results show that the proposal outperforms the best-of-breed algorithms commonly employed for spambot detection.