Again, I'm not saying it's a perfect solution. Perhaps a naive bayes classifier, with enough examples, could distinguish between negative opinions and hate speech. Maybe it couldn't -- it might be very similar to trying to create a sarcasm detector[1]. I'm hopeful to just get the ball rolling in terms of discussions -- even if I'm wrong, we are at least talking about ways to try to fix the problem.

[1] http://www.bbc.com/news/technology-27711109