One thing that annoys me about Mastodon is that the medium amplifies angry toots, particularly about politics. I don't want to read anything where anybody active after 1945 is accused to be "a Nazi" or "a fascist" for instance.
Because there are so many toots to read I have no problems losing some toots so if the model rejects some stuff because it is about a topic that people frequently write angry toots about that's fine with me.
My smart RSS reader YOShInOn is already classifying 2000 articles on a good day, one of these days I have to update it so it can train more than one classification model. I'm pretty sure my model would work OK for angry toots if I can collect 5000 or so them which might be a lot to bear.
An advantage of content-based filtering trained by the individual (as opposed to a corporation) is that it will not be so controversial because the people being filtered won't know that it happened. Someone on mastodon pointed out that this doesn't give backpressure for people to improve their behavior
https://mastodon.social/@UP8/110526769345483257
but