[thaumaturgy]

This seems like the ideal project for a ML experiment. Offer an integrated comments section, attach the ability for humans to moderate it, and then feed the moderated items into your favorite classifier for training and see if it can learn how to mute the louder nuisances without blocking more civil discussion.

Hand-moderating comments is a miserable job, it'd be a mercy if nobody had to do very much of it.