Figure 3
Evolution of COVID-19 content. Focusing on a single Gab hate cluster, this provides example output from our machine learning topic analysis. Although discussion of COVID-19 only arose in December 2019, it quickly evolved from featuring a large number of topics with a relatively low average coherence score, to featuring a small number of topics with high average coherence score more focused around COVID-19. As the right-hand panel shows, the discussion in this cluster became much more coherent, and focused on COVID-19, during the second 3-week period we analyzed. The right-hand panel shows the keywords in each of 5 topics discussed on this cluster during that second 3-week period. In the first three-week period, topics discussed featured profanity and hate speech such as f*** and n*****, but the conversation quickly become more focused and less like a stereotypical hate-speech rant.
