Twitch to use machine learning for spotting ban evaders

Twitch is advancing its efforts to decrease harassment with a new machine learning tool called Suspicious User Detection. The company will use it to detect people who might be attempting to avoid bans. It’s Twitch’s latest addition to fight against things such as hate attacks, where chats are overwhelmed with trolls spreading hateful messages.

The new tool can identify people who might be users who have avoided bans from a streamer’s channel. The machine learning prototype powering the tool distinguishes potential evaders. It does this by assessing their characteristics and behavior. Then, it compares that information against accounts banned from a streamer’s channel.

Mods and streamers can choose to monitor a possible ban violator. The program puts that user on a monitoring list and places a message next to a user’s name to remark on the monitoring process or ban them. Potential evaders’ messages will show up in the chat, but mods/streamers can also have those messages blocked from chat.

Suspicious User Detection (SUD)

The company says it will turn on SUD by default. However, streamers/mods can turn the tool off if they decide to or manually select to monitor suspicious users.

Twitch’s director of product for community health, Alison Huffman, said in a statement that community feedback inspired this tool for creating better ways to control ban evaders. 

She said that after speaking with mods about their issues, they discovered that it is difficult to distinguish whether a user who violated their channel’s norms was a repeat harasser or just a new viewer who hadn’t read the channel’s rules yet. She explained why they designed this tool to provide creators with further information about possible ban evaders to make informed decisions within their channels. SUD seems like it might make a difference in fighting against hateful individuals. But it needs time to check how effective it will be in practice.