This is pretty cool. Did you create the dataset (or is it available somewhere)?
The labelling takes it to the next level for auto-moderation. You could create a community based on "We're obscene, but we're not toxic or insulting". A lot of the datasets online are binary on a single level.