Pick as many active users whose judgment you trust as you can find, train a Bayesian classifier on their votes, up and down, and use that to score the voting patterns of users. Set ignore for the ones with the worst scores. Even if it turns out not to help much, at least you'll have had some fun doing it.

Also, there's currently nothing reminding users of the ideals you want them to uphold just before they submit a comment -- i.e., right next to the submit button. It never hurts to ask.