[klyrs]

I don't know if it does or doesn't, but Republicans and Democrats use different keywords, not just semantics. For example, 'law and order', 'police reform', etc. It's certainly possible for ML to pick up a bias in the presence of such differences.

[Retric]

They also quote each other, without context it’s just word choice not the message. Considering messages are already tailored to different kinds of voters, spam algorithms are just another part of messaging.