Keywords: online toxicity, online benevolence, Reddit, social media
When toxicity arises in the context of online discourse, how common is a benevolent response? We analyzed 10,963,849 comments posted to the discussion site Reddit in January 2016. Of these, 75,269 comments were classified as toxic using Perspective API on four toxicity metrics. A group of 792 online volunteers re-rated the toxicity of a random sample of 14,856 of these comments. They then read a direct reply to each comment and rated this reply on benevolence (e.g., how much the reply demonstrated understanding of the content, empathy for the commenter, and used a polite/helpful tone). We estimated that of the 8591 comments which were verified by our raters as toxic, 37.8% of replies were rated as above the scale midpoint in benevolence, suggesting that replying with understanding, empathy and politeness in the face of toxicity may not be as rare online as we might think. The average benevolence of these replies was 3.16 where 1 = not at all benevolent and 6 = very much (SD = 1.32). In future research, we plan to train a machine learning algorithm using these benevolence ratings with the hopes of classifying larger sets of toxic comment-reply pairs from Reddit, with the aim of further understanding the prevalence of benevolence in response to online toxicity.