Several people have mentioned to me that the comments on Hacker News seem much better recently, and asked me what changed. As far as I can tell, two of our experiments are working.
The first is posting feedback in the threads about what's good and bad for HN comments. Right now, dang is the only one doing this, but other moderators may in the future. We've learned some things about how to do it: 1) the feedback should be as neutral as possible; 2) it should be about the comment, not the commenter; and 3) where possible, it should say what would make the comment better.
Other HN users have been pitching in with feedback too, which is great! If you'd like to help, please do. Just try to follow the three guidelines above.
The second experiment is a change we made to the comment scoring and ranking algorithms. These algorithms do more than just counting and sorting because pg wrote a lot of code to address systemic issues as they came up over the years. But the community doesn't stay static, so the algorithms shouldn't either.
After studying the data, dang and kogir tuned the algorithms to make some downvotes more powerful. We've been monitoring the effects of this change, and it appears to be reducing toxic comments.
The majority of HN users are thoughtful and nice. It's clear from the data that they reliably downvote jerks and trolls (and specifically, they don’t silence minority groups—we’ve looked into this). What dang and kogir found was a way to turn the volume up on this kind of downvote. We believe this has made the comment scores and rankings better reflect the community.
We will be trying a lot more experiments. We'll stop the ones that don't work and continue the ones that do.
dang and kogir, great work so far. I'm enjoying reading HN much more.
It should be easier for a late-arriver on a post to add a useful comment, and have it be promoted. Have you considered using randomization to adjust the score of certain comments?
HN comments seem to exhibit a rich-get-richer phenomenon. One early comment that is highly rated can dominate the top of the thread. (I will note that, qualitatively, this doesn't seem as bad as a few months ago.)
The problem with this approach is that late commenters are less likely to be able to meaningfully contribute to a discussion, because their comment is likely to be buried.
One thing interesting about the way FB feed appears to work is that they use randomization to test the signal strength of new posts.
Have you considered using randomization in where to display a comment? By adding variation, you should be able to capture more information from voters about the proper eventual location for a comment. It also means more variation is presented to people who are monitoring a post's comments.