Twitch Leverages Machine Learning to Identify and Ban Trolls Dodging Rules

Twitch is rolling out a new machine learning feature, "Suspicious User Detection," designed to aid streamers in safeguarding their channels from users attempting to bypass bans. This innovative tool flags individuals deemed "likely" or "possible" ban dodgers.

For those identified as likely offenders, Twitch will automatically block their messages in chat and alert streamers and their moderators. This enables them to choose whether to enforce a ban. While possible repeat offenders can still send messages by default, they will also be flagged by the system. Streamers have the option to restrict these users from sending messages altogether.

Powered by a sophisticated machine learning model, Suspicious User Detection evaluates various signals such as user behavior and account characteristics, comparing this data against previously banned accounts to gauge the likelihood of evading a ban.

Although Twitch will enable this feature for all users, it will not automatically ban flagged individuals. The company emphasizes that streamers are the best judges of their own communities and should make the final decision regarding participant eligibility. As streamers engage with the system, its predictions will improve over time.

This new feature comes in response to recent challenges, including the surge of "hate raids," where malicious actors exploited bots to inundate channels with abusive language, particularly targeting creators from marginalized communities. The escalation of these incidents prompted some creators to temporarily leave the platform in protest of Twitch's insufficient response.