Understanding the Role of Worldviews in Identifying Online Toxic Conversations

Online public discussion has the potential to mitigate the disturbing trend of social polarization. However, as a result of the rise of online toxic behaviors, such benefits have been severely hindered. Based on the social judgment theory, this study examined the role of worldviews in predicting toxic behaviors. This work used toxic conversations on the Reddit changemyview subreddit and analyzed 45 million comments made by 9,546 users across all Reddit communities between 2015 and 2019. This paper found that, combining worldviews with basic text representation features (Word2Vec) on a simple logistic regression led the framework to outperform the state-of-the-art obtained on sophisticated deep learning models. The existence of toxic conversations could therefore be explained by worldviews. This work informs the design of toxicity detection systems to take into account not only the information extracted from current conversations, but also the worldviews shown in the utterances of users over the years