Empirical Study of Hate Speech in Social Media During COVID-19 Crisis in the United States

Abstract

As the COVID-19 outbreak, hate speech on social media towards Chinese and other Asian groups has encouraged "Sinophobia". To capture the situation of hate speech on Twitter, we hydrated the tweets discussing COVID-19, written in English language and have location within the USA. These tweets hydrated from Twitter API in span of 5 months (153 days). We have obtained 543,943 tweets in which we identify 40,579 Hate Speech occurrences. We categorized and analyzed them according to Pysentimiento model which is based on BERT models and, Latent Dirichlet Allocation Model (LDA). The results indicate that there are substantial associations between the increased amount of hate speech and the increased rate of deaths due to COVID-19,increased rate of new COVID-19 cases, and negative tests rate.