While online communities have become increasingly important over the years,the moderation of user-generated content is still performed mostly manually.Automating this task is an important step in reducing the financial costassociated with moderation, but the majority of automated approaches strictlybased on message content are highly vulnerable to intentional obfuscation. Inthis paper, we discuss methods for extracting conversational networks based onraw multi-participant chat logs, and we study the contribution of graphfeatures to a classification system that aims to determine if a given messageis abusive. The conversational graph-based system yields unexpectedly highperformance , with results comparable to those previously obtained with acontent-based approach.
展开▼
