I think a fundamental question is should an "Offensive Tweets" detector be global, or tuned to an individual user's sensibilities? Also what is the outcome met

I think a fundamental question is should an "Offensive Tweets" detector be global, or tuned to an individual user's sensibilities? Also what is the outcome metric an algorithm (or even a team of human reviewers) be optimized for?