Recognizing Explicit and Implicit Hate Speech Using a Weakly Supervised Two-path Bootstrapping Approach

In the wake of a polarizing election, social media is laden with hateful. To address various limitations of supervised hate speech methods including corpus bias and huge cost of annotation, we a weakly supervised two-path bootstrapping approach for an online hate detection model leveraging large-scale unlabeled data. This system outperforms hate speech detection systems that are trained in a manner using manually annotated data. Applying this model on a large of tweets collected before, after, and on election day reveals and patterns of inflammatory language.