How do you filter 'polite' from impolite comments? Sarcasm can be thinly veiled. Hateful comments can sound genuinely nice to a ML algorithm, but taken literall

How do you filter 'polite' from impolite comments? Sarcasm can be thinly veiled. Hateful comments can sound genuinely nice to a ML algorithm, but taken literally can paint people in a bad light. As for the real issue the article talks about, most misinformation comes from people bringing up things that nobody asked about. Nobody asked if the world is flat, nobody asked about vaccines. Nobody asks about race, but online video game lobbies are always harping on the issue. How do you filter out people who are clever enough trolls to game algorithmic filters?