Clive Thompson has a good article in Wired about some effective ways to tame comment trolls.
The four ideas he proposes are:
The standard approach:
Most told me that if you’ve got a high-volume site with political content—like Whitehouse.gov—you’ll also need to moderate postings by hand, hiring staff to look over each comment and delete the truly crazy hate-speech ones. The Huffington Post employs up to 25 people at a time to comb through its 35,000 comments a day.
The Slashdot Approach:
Slashdot has an automated system that randomly picks a handful of readers and gives them, for a day or so, the power to describe others’ comments with terms like “funny” or “off topic.” Those descriptions are translated into a score from -1 to 5. Readers can set their filters so they see only comments with high ratings—and trollery effectively vanishes.
The Disqus Approach:
In this paradigm, if a comment gets a lot of negative ratings, it goes invisible. No one can see it—except, crucially, the person who posted it. “So the troll just thinks that everyone has learned to ignore him, and he gets discouraged and goes away,” chuckles Disqus cofounder Daniel Ha.
The BoingBoing approach:
Whenever Nielsen Hayden encounters a nasty post—an ad hominem attack, for example—she leaves it up but removes all the vowels: y r fckng sshl, for example. The result is incoherent enough that it’s neutered, yet coherent enough that no one can cry censorship. The comment hasn’t vanished.
Best of all, because disemvoweling is visible, it trains the community. “You’re teaching the other commenters what the lines are by showing them comments that have stepped over the line,” Nielsen Hayden says.