Text Mining Domestic Extremism Topics on Multiple Social Media Platforms

The U.S. Capitol riot increased scholarly interest in the relationship between online platforms and domestic extremism. This study utilized text mining methods such as LDA topic modeling and toxicity analysis to assess the prominent narratives disseminated by influential far-right accounts on three social media platforms around the riot timeframe. We utilized a novel categorization system to classify topic streams into narrative frames. Parler posts contained the greatest conspiratorial content and cross-platform content sharing. Twitter was found to have a small number of far-right accounts that actively disseminate extremist views. YouTube far-right influencers produced the highest number of videos related to racial topics. Parler posts demonstrated a strong association between toxicity measures and current trending topics. A few mentions of far-right accounts within the top 20 topic streams, reveal that extreme racial and nationalist ideologies are active on the Twitter platform, even though overall toxicity was low. Our YouTube data was seeded by far-right accounts, and therefore, not surprisingly, the content was revealed to contain a high degree of racially-charged discourse. Our analysis revealed, however, that the YouTube data was also characterized by a high level of toxicity coming from both creators and commenters. Both Parler and YouTube demonstrated that some far-right content creators were able to drive up the toxicity of the discourse. These results attest to the impact social media influencers have on online discourse. Interestingly, however, the toxicity scores for related YouTube videos remained low, suggesting that the video-promotion algorithm does not necessarily increase exposure to radical content.