Early Prediction of Hate Speech Propagation

Online hate speech has disrupted the social connectedness in online communities and raises public safety concerns in our societies. Motivated by this rising issue, researchers have developed many machine learning and deep learning methods to detect hate speech in social media automatically. However, most of the existing automated solutions have focused on detecting hate speech in a single post, neglecting the network and information propagation effects of social media platforms. Ideally, the content moderators would want to identify the hateful posts and monitor posts and threads that are likely to induce hate. This paper aims to address this research gap by defining a new problem of early hate speech propagation prediction. We also propose HEAR, which is a deep learning model that utilizes a post's semantic, propagation structure, and temporal features to predict hateful propagation in social media. Through extensive experiments on two publicly available large Twitter datasets, we demonstrate HEAR's ability to outperform the state-of-the-art baselines in the early prediction of hateful propagation task. Specifically, with just 15 minutes of observation on a post's propagation, HEAR outperforms the best baselines by more than 10% (F1 score) in predicting the eventual amount of hateful posts it will induce.