I suspect that it's now a solvable problem to determine if the first few messages from an account are, if not "harassment", at the least... argumentative. Angry. Uncharitable. Sentiment analysis is pretty well-established, as is figuring out what people are talking about for the purpose of advertising at them.

If a new account joins and immediately starts expressing strong opinions on, say, a certain Breitbart reporter and how much ethics is in his journalism, whatever those opinions are, those don't need to be sent as notifications or show up as a reply to other tweets. It's not like Twitter guarantees reliable delivery, anyway. If they've been around on Twitter for a bit, sure.

But it's not as clearly in the interest of profit as identifying advertising interests (or even identifying spammers).