Replying to @deaneckles
The problem I’m hinting at with that kind of optimisation is that it really does feed into explosive models. If angry people are likely to share something that angers them (maybe for specific types of anger) then they are spreading content likely to anger others.