Leaving off the problem of negative/angry/.... comments, is the current state of sentiment analysis or AI sophisticated enough to determine that a set of comments are higher quality?
"Those other people should die in a fire"
"But you should shoot them too, just in case"
These might be funny, light-hearted, sarcastic, and quality in some topics and angry, mean, toxic in others.
I'm sure it is being worked on but I don't see it happening anytime soon. It seems like it would require to much training and that the biases of the trained data would still produce incorrect outcomes.