Twitch is trying to deploy a new tool called Suspicious User Detection. This is intended to be done exactly as it appears in the box. Try to detect suspicious users. More specifically, those who are trying to circumvent existing bans.
of Announcement blog, Twitch’s comment:
When you ban someone from your channel, they should be banned permanently from your community. Unfortunately, malicious people often choose to create a new account, return to chat, and continue their abusive behavior. Detecting suspicious users using machine learning helps identify those users based on the number of account signals. By detecting and analyzing these signals, the tool can flag suspicious accounts as “potential” or “potential” channel ban avoiders and take action as needed. To.
What happens after a user is flagged depends on the flag. Users suspected of being “likely” will not see the message in a typical chat, but will continue to see it in mods and creators, and can take further action on the user.
On the other hand, in the “possible” case, you can post a message to the chat and see the message appear, but the author / modifier will “watch the user and if necessary” to these messages. Will be flagged so that you can limit chat.
Suspicious user detection is turned on by default when live, but streamers can go to settings and do things like increase posting limits for “potential” cases, add users manually, and so on. I can. Machine learning hasn’t been detected, but it’s suspicious anyway.
This all sounds great, but it also puts some very powerful tools in the hands of the machine. This is something that Twitch should be aware of by adding:
One thing to prepare, especially before and after launch, is that machine learning will never be 100% accurate. This means that there are false positives and possible false positives. Therefore, suspicious user detection does not automatically ban all possible or possible evaders. You are an expert on your community and need to make a final call on who can participate. The tool will learn from the actions it takes, and as a result, the accuracy of the predictions should improve over time.
Watch this video to see how the system works.