Dang, already? Sounds like Arc is working out well.
Still, though, any statstical approach (which is what I assume you're using) is going to have some lag because you have to wait for a statistically-significant sample. You can catch any given offender after he's caused an O(1) amount of trouble, but if the offenders don't correlate with each other, you can get overwhelmed by a sufficiently large number of them. Spammers certainly do correlate with each other, but it's not obvious that voting rings do.
I'm prone to think that pg's algorithm is, to some extent, content based. I have no real evidence to support this, just the evangelism in "A Plan for Spam."
Right. That's what I meant by "spammers correlate with each other". I'd be very surprised if content-based filtering were useful for catching voting rings. You can use graph-theoretic metrics to catch them, but unlike spammers, identifying one doesn't obviously help you identify others.
