Twitch is reportedly stepping up its efforts to curtail harassment, with a new tool powered by machine learning to identify people who could be trying to avoid bans. It is the company’s most recent addition for dealing with issues like hate raids, where streamer chats are loaded with trolls sending hate texts.

The new tool is named ‘Suspicious User Detection’ and can identify users into two categories, namely “possible” or “likely” people who have dodged bans from channels. The tool, operating with machine learning, identifies potential evaders by assessing their behavior and account features and compares the information with that of already banned accounts.

Messages from “likely” evaders will not go to chats, but streamers and their mods may see them. They can then decide to keep a check on a likely ban evader. This will directly ban the user, or add them to a monitoring list, marking the user’s name as being monitored.

On the other hand, texts from users classified as "possible" evaders, will be displayed in the chat, but streamers and mods can choose to block those messages as well.

Twitch mentions that it will switch on Suspicious User Detection by default, but streamers can change or disable the tool if they want. Streamers and mods will also be able to manually choose to monitor users they suspect.

Twitch’s director of product for community health, Alison Huffman, mentioned that the tool has been inspired significantly by community feedback that demanded better ways to curtail ban evaders.

Notably, the Suspicious User Detection tool is a promising prospect for silencing haters, particularly if used together with the recently introduced controls that allow streamers to require a phone or email authentication of accounts participating in the chat.

However, the effectiveness of the Suspicious User Detection is yet to be tested in practice to check whether ban evaders will come up with ways to sidestep the tool.

Source Credits –

https://www.theverge.com/2021/11/30/22810180/twitch-machine-learning-suspicious-user-detection-bans