Is "best" really the proper defining word to use? More eyeballs are viewing the RIP Michael Jackson thread, thus resulting in more potential up-votes -- based on that, I'd be hard-pressed to establish it as the "best" comment... maybe "most up-voted" is more fitting.
> More eyeballs are viewing the RIP Michael Jackson thread, thus resulting in more potential up-votes
That suggests an idea to take that into account, but I think PG said he'd rather not complicate the algorithm.
Edit: Huh? I'm being downvoted, which is fine, but why? I meant that there might be an interesting way to take eyeballs into account to weight a vote, but IIRC, PG wanted to see how long we could go without vote weighting.
This is because many people here still vote by agreement, rather than for comments that are interesting, insightful, contradictory, terse, and so on. This happens on a much larger scale with lower-rated comments. I once saw the word "no" with a score of 50~, which made up for the contributor's running streak of -5~. It wasn't that their contribution quality became better, it was that their comment hit that major "upvote when you agree" fault in our system. It's infuriating and seriously reduces our quality. I try to upvote even poorly composed devil's advocate comments just to counteract this, because even reckless disagreement is often a gold mine of those little bits of information that you didn't take into consideration, but should have.
Vote for what you think others should see, not for what you agree with.
many people here still vote by agreement, rather than for comments that are interesting, insightful, contradictory, terse, and so on
Given up and down arrows next to a comment, the overwhelming majority of people will always vote based on agreement, no matter what you tell them to do, because the user interface practically begs people to do that.
If you want people to vote based on something else, you'll have to redesign the user interface (and even then I suspect it will be difficult).
It's not that you should avoid upvoting what you agree with, just that it shouldn't be the only reason you upvote. A lot of the posts that violate this simply bait existing divisions, which contributes nothing. Perhaps I should add "and people don't need to see that which they already agree with."
It's a community attempting to define its norms. I'd rather have "this is not an appropriate post for this site" upvoted to the skies than nearly silent removals of content deemed unacceptable or a ten page "me too!" AOLgasm.
It seems that most of the highly voted comments are about how something/someone is bad/wrong, but there are some positive ones in there too, like those by ktharavaad, lutorm, kirse, patio11 (x3), jbr, plinkplonk, chriseppstein. Actually, 11 out of 30 isn't bad! I tend to forget that Hacknews is a public forum on the internet.
I think I just need to focus on the worthwhile comments, and ignore the negative ones (but it's hard: http://xkcd.com/386/). If everyone did this, the problem wouldn't exist.
BTW: On the homepage (not this one), see the "lists" link at the very bottom, on the left. It's: http://news.ycombinator.com/lists The 3rd one, "bestcomments" is what been submitted here, but with an argument "?newsToMe" appended (presumably to avoid hitting the last story about it, 544 days ago: http://news.ycombinator.com/item?id=93526)
Agreed, this would be closer to a correct outcome. Especially for users like me who generally comment in offpeak times (being on the other side of the world) and usually come to a discussion many many hours after it has long died down, so his comments remain at the bottom of the page, ignored.
I'd assume it works on the same algorithm as the home page and normal comment pages. Something of the form:
points / (timeSinceEpoch^2)
After a certain time the number of points needed to override the denominator is unattainable, no matter how popular the comment was. As a result, the comments on the top of this list should be both popular and recent.
It's the latter. A comment with 30 points, say, a year ago would have been very unusual, but now that's a pretty typical score for the highest-rated comment of any link in the current top 10.
Unfortunately, this seems to coincide with HN following the same trajectory as early Reddit, and that makes me sad.
While I do share your sentiment fearing that the community will change, I don't think that the karma totals are really a good gauge of content/comment quality on HN. Just because more people are here to upvote does not mean the comment is of lesser quality than before (the opposite is also true...)
A quick fix to this could be a sort of weighted comment score. Maybe the karma points / by the number of karma points added that day to give an idea of scale in comparison to the overall size of community. I'm sure others could come up with something more precise, but it's a start.
I don't like capping, sometimes there are truly exceptional comments that deserve 99.9%ile rankings. What about some form of logarithmic decay? Perhaps the first upvote should not be worth the same as the 101st.
By the same token, it seems to me that the ratio "story upvotes"/"number of comments" has gone higher and higher recently.
I'm not sure what it says about HN though. I might not necessarily be a bad thing actually. I'm not really for having a flood of comments on all the stories.
I know this post is about the "best" comments, but that doesn't necessarily indicate quality. I'm not sure what might be a good algorithm to judge quality, but popularity is most certainly not it. To all you natural language processing nerds out there...maybe there is a way of looking at word choice, with weights also given to amount of karma but relative to the other karma distributed to other comments.
This is insane, not because you're wrong, but because popularity and quality should be equivalent for us. We have some of the very best natural language processing algorithms going over most of our posts - and several comments in this post are suggesting that they're failing miserably (in many important cases, I agree).
This is a problem with the community, not with the software.
One thing I dislike about the simple arrows is that they give no indication of on what basis posts should be upvoted. If, when clicking, we were forced to select from a menu vote types [+1 insightful, +1 clever, +1 informative, +1 antigroupthink, +0 I agree, +0 funny, -1 rude or troll], even if we did not track these, this would be giving people feedback on voting expectations and force them to pause for one second to put a bit more consideration into their vote.
"About time... I can't be the only one who thinks Experts-Exchange is lame..." - I don't mean to impugn the author of that comment, but I think the amount of approval for it indicates that some people are voting on agreement more than quality.
I'd like to see karma points per view. As there are more and more people on HN, the views go up, and hence the upvotes. More, any comment posted at an unpopular time gets overlooked and sinks without a trace. Having karma per view would help to offset that problem, if not fix it.
Currently the time factor is doing a similar job, but I think page views would be a better value to use.
