This comment seemed to me a perfect specimen of the type that drags down forums: vitriolic and ill-informed. I've hypothesized before that the problem is not that people make such comments (which seems inevitable if you have open, anonymous signups) but that others upvote them. So I analyzed the votes to see if there was a pattern, and indeed there is. The median karma of upvoters was 644, and the median karma of downvoters was 1814, almost 3x as high. If this pattern holds up it could be very useful.
What does the ratio look like on a "neutrally controversial" comment look like? That is, a comment that isn't toxic, but merely dividing the community.
Correct me if I'm wrong, but if that pattern is sustainably observed, then it almost certainly means the upvote/downvote privileges should be changed, correct?
Elsewise, the pattern will continue like entropy, until the majority of comments are like that.
There already are certain privileges to karma: you need a certain threshold to downvote comments at all, and I suspect that thresholds to get onto the front page are based on the karma of who upvotes your story. (There've been times where my upvote alone, going from 2 to 3 votes, is enough to get something on the front page, while many times I've seen new stories with 4-5 points that don't make it.) I suspect PG is thinking of tweaking the weights, automatically, until there's enough of a weight to high-karma downvotes to kill bad comments. You could easily train an SVM or other classifier on some set good/bad comments and then pull out the coefficients to figure out what signals should go into weighting point totals.
Karma doesn't necessarily translate to expertise. Most of my highest-karma comments have been utter nonsense. High karma can mean people have agreed with you, or that you've been here longer than not, and it probably is a good measure that you can conform to the norms of the community, but it doesn't necessarily mean you know what you're talking about.
I mean, people will listen when tptacek talks about cryptography, but that doesn't mean his opinion on design should necessarily matter more than that of a designer with less karma just because tptacek has the bigger integer attached to his name. It's not really more 'fair' for votes by high karma users to outweigh votes by lower karma users.
Unless it already works that way in which case it's the best thing ever and clearly pg is going God's work.
I've worked on content-based Bayesian voting schemes. They work well weighting votes by content words. tptacek + cryptography should matter more.
Here, though, it seems as if HN experts and novices may be representative. pg used to run the front page based on karma. Comments seems like a place where that simplistic approach could work well.
