Predicting the Hate: A GSTM Model based on COVID-19 Hate Speech Datasets

COVID-19 crisis has been accompanied by copious hate speeches widespread on social media. It reinforces the fragmentation of the world, resulting in more significant racial discrimination and distrust between people, leading to crimes, and injuring individuals spiritually or physically. Hate speech is hard to crack for a global recovery in the post-epidemic era. Conducting with Twitter datasets, this paper aims to find the key indicators that influence the trend of hate speech, then builds a Gaussian Spatio-Temporal Mixture (GSTM) model for trends prediction based on the pre-analysis. Findings show that in the early period, the participation of influential users is closely related to the emergence of sentiment peaks, and the interval time is around one week. After hate speech waves up, the indicator of total exposure becomes more critical, suggesting that grass-root release influences at this stage. Compared with three classical time-series predicting models, the GSTM model shows better peak prediction ability and lower residual mean. This work enriches the approaches of predicting unknown but foreseeable hate speeches accompanied by future pandemics.