zlacker

>>pg+(OP)
3 words: Bayesian Comment Filter. Just does the opposite of what the spam filter does. Use the corpus of great comments from the past to find great comments of the present.

I'm only half joking. Fundamentally, the thread is about a filtering system.

>>b_emer+A4
That could work, actually. Instead of focusing on discouraging bad comments, maybe the answer is to promote good ones. Plus I have some code I could use for that.

>>pg+65
But it raises the question, do you have a way to single out the corpus of good comments from the past? If past up-votes were reliable I don't think this thread would exist.

>>colins+Ka
I could train it on stuff from "the good old days."

>>pg+lk
What if I get a bunch of good comments from the good old days, train a Bayesian filter on them, and then make a comment bot with bias in my favour?

Perhaps you could give everyone a comment bot, like a green/red bar that says whether the comment looks like low quality or high quality as you're typing it. A lot of people might edit the comment to make it better, or simply delete the comment (you could design UI to encourage this ... eg RBM's can highlight which words look like they're causing the problem, or offer a Kill Comment "X" when the comment is far into the red).

You could also train the bayesian filter on (graphwise) voting patterns, rather than on comments as bags of words.