The effects of user moderation in social media

For many years, content moderation on social media was restricted to deletion; users were only able to report content for manual review thus creating a large workload for human moderators. However, recent developments in the field of artificial intelligence open up possibilities to go beyond this traditional approach: AI moderators can be trained to act on their own by first determining when the deliberative quality starts to deteriorate and then to decide on the type of intervention needed. In this paper, we are laying the foundation for the latter direction of research: we conduct a large-scale analysis of X and Reddit posts with the question of how deliberative quality improves when users intervene. Given that a user has intervened and pointed out a misconduct of a deliberative norm, for instance by telling other users to be more friendly, we analyze the reactions towards these interventions. For a selected set of deliberative norms (and violations thereof), we develop computational features allowing a large-scale analysis of the conversations and users. Based on our results, we suggest a set of (personalized) interventions that are most likely to succeed in improving inappropriate deliberative online behavior.