Twitch Battles Ban Evasion using Machine Learning

| | , , ,

Twitch streamers have had a hard time with hate raids and ban evasion recently. Back in September, Twitch launched a verified chat in which viewers must verify via phone or email before they are able to chat. They put this into place to help keep hate and harassment out of channels. This has helped minimally, and Twitch has acknowledged that they still have more to do to ensure they provide a safe space for everyone.

Twitch addresses harassment

Twitch has picked up on the need to introduce suspicious user detection. This is a tool to help creators and moderators catch users attempting to evade channel bans.

The company believes that when you ban a user, they should be banned for good. They have noticed that there has been an ongoing issue with ban evasion and have been working hard to address the problem. Twitch announced that ‘Bad actors often choose to create new accounts, jump back into chat and continue their abusive behavior’. They have introduced Suspicious User Detection, powered by machine learning, to help identify those users based on a number of signals.

‘When you ban someone from your channel, they should be banned from your community for good’

– Twitch

Suspicious User Detection will detect and analyze these signals then flag the suspicious account. The account will be flagged as ‘likely’ or ‘possible’ channel ban evaders allowing you to take action.

How will the accounts be flagged?

For likely ban evaders, messages will not appear in chat. However, the messages sent will be visible to creators and moderators. They will be given the choice to leave the restriction and monitor the user, or ban them from the channel.

‘Possible’ ban evaders will be able to chat normally. However, they will be flagged to the creator and moderators. This allows them to monitor the user before deciding whether or not restricting them is necessary.

If you’d like to be extra cautious, you’re able to adjust the settings for both groups. For example, you’re able to restrict both likely and possible ban evaders from appearing in the chat. You may also deactivate the setting altogether via the moderation settings page. It’s also possible to manually add users that you’d like to monitor, either from the new suspicious user’s widget in Mod View or by selecting the user’s viewer card.

Can we expect this to end ban evasion?

No machine is 100% accurate. There’s going to be a possibility of false positives and negatives, particularly around launch time. Twitch states that ‘you’re the expert when it comes to your community. That’s why suspicious user detection doesn’t automatically ban all possible or likely evaders. With all machine learning technology, the tool will learn from the actions you take. This will improve the accuracy of its predictions over time.

[ header image ]

Zoe Craddock

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.