Twitch gets AutoMod with language recognition and machine learning

Twitch is getting a so-called AutoMod. It should be able to automatically detect different types of inappropriate messages and temporarily block them until a human moderator has been able to approve them.

The language recognition of AutoMod works according to Twitch based on the same techniques as Siri, Google Assistant and Amazon Echo. The company combines that with machine learning to get better at identifying inappropriate statements over time. AutoMod can be set to four different ‘modes’ for its controls and divides statements into four categories: identity, sexually explicit language, aggressive language and swear words.

Users who harass each other and streamers have long been a problem on major platforms such as Twitch. While a moderator is an appropriate solution in principle, their job becomes more difficult as more users become hostile in chat. AutoMod should make pre-approving posts by mods manageable by setting aside only the offensive posts for approval.

According to PC Gamer, AutoMod has been used successfully before. The bot would have been used during the live streams of the National Conventions of both the American Democratic and Republican parties. “The civilized way Chat interacted with each other earned compliments,” said Twitch. AutoMod now works in English, but beta versions are available for Arabic, Czech, German, Italian, Japanese, Korean, Polish, Portuguese, Russian, Spanish, and Turkish.