A Unified Deep Learning Architecture for Abuse Detection.

Hate speech, offensive language, sexism, racism and other types of abusivebehavior have become a common phenomenon in many online social media platforms.In recent years, such diverse abusive behaviors have been manifesting withincreased frequency and levels of intensity. This is due to the openness andwillingness of popular media platforms, such as Twitter and Facebook, to hostcontent of sensitive or controversial topics. However, these platforms have notadequately addressed the problem of online abusive behavior, and theirresponsiveness to the effective detection and blocking of such inappropriatebehavior remains limited.

In the present paper, we study this complex problem by following a moreholistic approach, which considers the various aspects of abusive behavior. Tomake the approach tangible, we focus on Twitter data and analyze user andtextual properties from different angles of abusive posting behavior. Wepropose a deep learning architecture, which utilizes a wide variety ofavailable metadata, and combines it with automatically-extracted hiddenpatterns within the text of the tweets, to detect multiple abusive behavioralnorms which are highly inter-related. We apply this unified architecture in aseamless, transparent fashion to detect different types of abusive behavior(hate speech, sexism vs. racism, bullying, sarcasm, etc.) without the need forany tuning of the model architecture for each task. We test the proposedapproach with multiple datasets addressing different and multiple abusivebehaviors on Twitter. Our results demonstrate that it largely outperforms thestate-of-art methods (between 21 and 45\% improvement in AUC, depending on thedataset).