Detecting Flagged Comments by Analyzing User Behavior Features in Online Communities

Comments violating community guidelines have long been a challenge in online discussion communities, often disrupting the user experience and overall community health. This study presents a quantitative analysis of such flagged comments on a large Chinese online discussion platform, investigating factors including comment deletion rates, sentiment, discussion context, user voting behavior, and publication timing. Drawing on a dataset of 6,900,119 historical comments from 17,226 users, which were carefully collected and cleaned, we developed a predictive model based on these behavioral features. The model demonstrated high accuracy in identifying flagged comments, achieving an AUC of 99.2%.