Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Hacker News metrics (first rough approach)

mjg59.dreamwidth.org

Post reply on HN
Print view View original

31–40 of 72 posts

Re: Hacker News metrics (first rough approach)

#31

Post by waterlesscloud »

JoshTriplett wrote: post #29
slg wrote: post #24

Earlier quoted context omitted.

I seem to remember reading that in an attempt to avoid "controversial" subjects, posts are penalized if they generate comments too quickly. Social issue are often hotly debated and occasionally some of the most actively commented on posts on the site until they drop off the front page. So while HN might not actively be penalizing these posts, they might be penalized in indirect but still very real ways.

I would hope that HN's "flamewar" detector is somewhat more sophisticated than just "generate comments too quickly".

There's an aspect of it that's a ratio of comments to upvotes, which seems to work pretty well from what I've observed.

Re: Hacker News metrics (first rough approach)

#32

Post by JoshTriplett »

waterlesscloud wrote: post #31
JoshTriplett wrote: post #29

Earlier quoted context omitted.

I would hope that HN's "flamewar" detector is somewhat more sophisticated than just "generate comments too quickly".

There's an aspect of it that's a ratio of comments to upvotes, which seems to work pretty well from what I've observed.

Now that makes more sense. A story that generates a large number of comments and a large number of upvotes very quickly seems likely to be a good story. A story that generates many quick comments without upvotes seems likely to have produced comments negative about the article; if you posting a comment but you don't want to upvote the article, that does seem like a good sign of controversy and possible flames.

Re: Hacker News metrics (first rough approach)

#33

Post by dang »

nknighthb wrote: post #27
dang wrote: post #10

The Hacker News ranking algorithm tends to penalize stories that address social issues. The Hacker News ranking algorithm does not penalize stories that address social issues. It doesn't consider social issues at all, apart from a couple of special cases that we turn off once a major wave subsides [1]. The effects the OP is observing are mostly caused by user flags and the voting ring detector. 1. The intention there…

This is insanely disingenuous. User flags and the voting ring detector are part of HN's ranking algorithm, whether they're labeled as such in the code or not.

We must have a crossed signal here. Of course those are part of the ranking algorithm! My point is that except for the exceptions I mentioned, none of the code considers social issues. What part is disingenuous?

Re: Hacker News metrics (first rough approach)

#34

Post by minimaxir »

dang wrote: post #21
somehow wrote: post #16

Earlier quoted context omitted.

I would be more fair to address the broader point of the post rather than just one sentence. It's good that the moderator(s) specify why something is penalized[0], but it doesn't really change the overall result. [0] As defined by the post in question.

Stories about hot social and political controversies reliably attract both a lot of upvotes and a lot of user flags. This, together with how the software interprets votes and flags, explains the effects the OP is describing. On hot social and political issues, many people find it irresistible to conclude that HN is biased against the side they favor and favors the side they oppose. They perceive this bias in all thin…

You may want to consider writing a blog.ycombinator.com post about this. I think there are lots of HNers who would find these algorithms interesting :)

Re: Hacker News metrics (first rough approach)

#35

Post by slg »

JoshTriplett wrote: post #32
waterlesscloud wrote: post #31

Earlier quoted context omitted.

There's an aspect of it that's a ratio of comments to upvotes, which seems to work pretty well from what I've observed.

Now that makes more sense. A story that generates a large number of comments and a large number of upvotes very quickly seems likely to be a good story. A story that generates many quick comments without upvotes seems likely to have produced comments negative about the article; if you posting a comment but you don't want to upvote the article, that does seem like a good sign of controversy and possible flames.

But it doesn't just penalize flamewars, it penalizes all conversations. For example, I have now posted two comments in this discussion and only upvoted the article once. Does that mean this is a worse article than one that I only upvoted and never commented on or a better article than one I upvoted and commented on three times?

Re: Hacker News metrics (first rough approach)

#36

Post by krapp »

JoshTriplett wrote: post #32
waterlesscloud wrote: post #31

Earlier quoted context omitted.

There's an aspect of it that's a ratio of comments to upvotes, which seems to work pretty well from what I've observed.

Now that makes more sense. A story that generates a large number of comments and a large number of upvotes very quickly seems likely to be a good story. A story that generates many quick comments without upvotes seems likely to have produced comments negative about the article; if you posting a comment but you don't want to upvote the article, that does seem like a good sign of controversy and possible flames.

I think it's mistaken to assume that not upvoting and downvoting are each implicitly negative, though. There is a neutral dimension to subthreads as well - conversations which are neither praiseworthy nor trolling.

Re: Hacker News metrics (first rough approach)

#37

Post by Chinjut »

krapp wrote: post #30
slg wrote: post #24

Earlier quoted context omitted.

I seem to remember reading that in an attempt to avoid "controversial" subjects, posts are penalized if they generate comments too quickly. Social issue are often hotly debated and occasionally some of the most actively commented on posts on the site until they drop off the front page. So while HN might not actively be penalizing these posts, they might be penalized in indirect but still very real ways.

I believe a thread with more comments than votes is penalized as well, suggesting that commenting on a thread without upvoting it first is implicitly flagging it. And of course there may be other arcane metrics involving thread depth, average comment rate, comment size, thread shape, etc. Who knows?

Whoa, really?

...I've been flagging a lot of articles.

Re: Hacker News metrics (first rough approach)

#39

Post by mjg59 »

dang wrote: post #33
nknighthb wrote: post #27

Earlier quoted context omitted.

This is insanely disingenuous. User flags and the voting ring detector are part of HN's ranking algorithm, whether they're labeled as such in the code or not.

We must have a crossed signal here. Of course those are part of the ranking algorithm! My point is that except for the exceptions I mentioned, none of the code considers social issues. What part is disingenuous?

If social issues are more likely than the average story to trigger aspects of the ranking system that results in them disappearing, the ranking system is biased against social issues. The fact that it's also biased against various other topics is irrelevant to the point I was making - you've constructed a ranking system that makes it almost impossible to maintain visibility of stories about diversity. In an era where the lack of diversity in our industry is so extreme and stories about harassment so visible that the mainstream media have picked up on it, that seems like a problem.

Re: Hacker News metrics (first rough approach)

#40

Post by waterlesscloud »

mjg59 wrote: post #39
dang wrote: post #33

Earlier quoted context omitted.

We must have a crossed signal here. Of course those are part of the ranking algorithm! My point is that except for the exceptions I mentioned, none of the code considers social issues. What part is disingenuous?

If social issues are more likely than the average story to trigger aspects of the ranking system that results in them disappearing, the ranking system is biased against social issues. The fact that it's also biased against various other topics is irrelevant to the point I was making - you've constructed a ranking system that makes it almost impossible to maintain visibility of stories about diversity. In an era where…

Given how often diversity stories are on the front page, there doesn't seem to be much of a problem getting such stories visibility.

In fact, the top story all day long today has been a diversity story, with a massive number of comments on just about every possible aspect of the story.

Post reply on HN