Twitch Machine Learning Will Catch Ban-Evading Users
In a bid to curb abusive behaviour and harassment, Twitch will start using machine learning to detect users trying to evade bans. Get the full details here.
A streaming platform for sharing live video content, and a community for many gamers, Twitch has not been without its problems. As with most sites that enable user interaction, sadly the chat feature on Twitch has been used as a tool for spreading hate on the platform. Not dissimilar to the dislike button on YouTube being utilised as a weapon, which has now been hidden from view.
Raids on Twitch are designed to promote newer channels and send viewers over to other streams as one is ending, via the chat feature. “Hate raids” have become commonplace, with abusive users flooding the chat windows of targeted channels.
Whilst there are some measures in place to detect and deal with this behaviour, it isn’t a flawless system. Users and bots created for taking part in hate raids can use numbers and special characters to disguise banned language.
As a result, victimised channels are still objected to abuse. Abuse that is visible to all others viewing the stream, causing stress and embarrassment. This is particularly upsetting for streamers aiming to launch careers in Twitch streaming.
How Will New Machine Learning Work?
A new tool, called Suspicious User Detection, will be turned on by default for all channels when it goes live. Announcing the move in a blog article, Twitch stated:
“When you ban someone from your channel, they should be banned from your community for good. Unfortunately, bad actors often choose to create new accounts, jump back into Chat, and continue their abusive behavior. Suspicious User Detection, powered by machine learning, is here to help you identify those users based on a number of account signals. By detecting and analyzing these signals, this tool will flag suspicious accounts as either “likely” or “possible” channel-ban evaders, so you can take action as needed.”
The AI will sort detected users into “likely” and “possible” ban evaders. Different approaches are taken towards each category.
Likely evaders’ messages will not be sent to chat. Although, their messages will be visible to creators and moderators of the channel. This will allow mods to monitor that user, or ban them from the channel.
Possible evaders’ messages will be sent to chat for all to see, however the account will be flagged so as to notify creators and mods. From here, mods can monitor the user and restrict them from the chat if they wish.
Twitch has highlighted the fact that no machine learning is 100% accurate when first introduced. Creators should expect a certain degree of teething problems. As such, Twitch has designed these measures so that the ultimate decision on banning lies with the creator.