Content Moderation of Speech in Political Discussions

Social media platforms have been hailed as “politically disruptive communication technologies” (Hong & Nadler, 2012). Individuals express opinions and engage with politicians, the press, and each other on social media, sometimes using offensive language (Rossini et al., 2020). Content moderation has been adopted by many social media platforms to screen and evaluate offensive speech. In the present study we trained offensive speech classifiers to analyze offensive speech examples by integrating three archival datasets. We then used the trained classifier to examine a large body of comments about YouTube videos posted during the 2018 midterm election cycle. This provided information on the prevalence of various kinds of offensive comments and the pattern of content moderation used by YouTube. We also examined comment negativity using offensive speech lexicons. Our results showed systematic variance in the prevalence of offensive speech topics depending upon the political orientation of the content. Language use was significantly different between left and right‐leaning videos for comments related to sexism.