"You Know What to Do"

• DOI: 10.1145/3359309
• Publication Date: 2019-11-08T13:40:21Z
• Authors: Enrico Mariconti; Guillermo Suarez-...; Jeremy Blackburn; Emiliano De Crist...; Nicolas Kourtellis; Ilias Leontiadis; Jordi Luque Serrano; Gianluca Stringhini
• Keywords: Phishing; Relevance; Thumbnail

Video sharing platforms like YouTube are increasingly targeted by aggression and hate attacks. Prior work has shown how these attacks often take place as a result of "raids," i.e., organized efforts ad-hoc mobs coordinating from third-party communities. Despite the increasing relevance this phenomenon, however, online services lack effective countermeasures to mitigate it. Unlike well-studied problems spam phishing, coordinated aggressive behavior both targets is perpetrated humans, making defense mechanisms that look for automated activity unsuitable. Therefore, de-facto solution reactively rely on user reports human moderation. In paper, we propose an identify videos likely be harassers fringe communities 4chan. First, characterize model along several axes (metadata, audio transcripts, thumbnails) based ground truth dataset were raids. Then, use ensemble classifiers determine likelihood video will raided with very good results (AUC up 94%). Overall, our provides important first step towards deploying proactive systems detect YouTube.