[DanHulton]

From the article: “I'd emphasize that our technology is based on account behavior not the content of Tweets.”

It sounds like it's an automated system attempting to de-emphasize certain uses of the platform, not someone standing there going "Okay, Republican? Shadowbanned. Next? Republican? Shadowbanned. Next?"

i.e. if you don't want to be shadowbanned, maybe stop acting poorly.

[ZainRiz]

Most likely this is an AI/ML bases system that is doing Natural Language Processing to determine if an account is exhibiting "bad behavior"

That means first they identified many badly behaving accounts, trained their ML model to auto-detect other accounts behaving badly, and then banned them.

Of course, these ML models are famously hard to decipher.

The implication?

Twitter could choose to mention the criteria they used to select the original set of accounts they used to train their model (most likely it was accounts that had been manually banned in the past, these would have been bans which went unnoticed by the media) but they can't say "our model auto-bans accounts which do X, Y, and Z" because the logic used by the ML models is too obscure for any person to understand.

[mcphage]

> Twitter could choose to mention the criteria they used to select the original set of accounts they used to train their model

The first step to blocking trolls is to not help them understand how they got identified. Seriously, ask Google what algorithm they use to block scammers. They aren’t going to tell you, because then the scammers will just use it to screw with search results even more.

[mkempe]

So, a group of notorious Republican congressmen are trolls and scammers? that's how you believe Twitter makes decisions?