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

Hacking the Hivemind: Predicting Comment Karma on Internet Forums (2014) [pdf]

cs229.stanford.edu

Post reply on HN
Print view View original

31–40 of 52 posts

Re: Hacking the Hivemind: Predicting Comment Karma on Internet Forums (2014) [pdf]

#31

Post by 3131s »

blazespin wrote: post #14
mrlyc wrote: post #3

That article reminds me of The Auditors in Terry Pratchett's "The Thief of Time" who try to analyse a painting by reducing it to its component molecules. There is no mention of tailoring comments and replies to the target audience. Once I adopted the mindset of a senior high school or college student and posted accordingly, I was able to increase my karma on Reddit from 1,000 per year to 1,000 per month. I still have…

Does HN even respect karma? My guess is no and is just there as a red herring(not necessarily intentionally) that keeps people from cracking the real metrics it's using.

What do you mean by respect karma? Obviously the ordering of sibling comments is determined in some way by upvotes and downvotes.

Re: Hacking the Hivemind: Predicting Comment Karma on Internet Forums (2014) [pdf]

#32

Post by 3131s »

Xcelerate wrote: post #11
rosalinekarr wrote: post #4

Earlier quoted context omitted.

> I still haven't managed to crack Hacker News, though. Tough audience. Just link a bunch of studies to anything you say.[^1] It doesn't actually matter if they're related.[^2] No one really checks the sources anyway.[^3] [^1]: https://arxiv.org/abs/1502.02876 [^2]: https://arxiv.org/abs/1610.02728 [^3]: https://arxiv.org/abs/1611.03799

Haha no way. My highest rated posts have all been some vapid feel-good thing that I posted shortly after an article was submitted to HN. My detailed explanations that link to studies rarely get more than 10 points (same phenomenon occurs on Quora as well actually...)

I always wonder how much these factors affect a comment's score:

* Proximity to the top of the page

* Time of day posted

* Number of sibling comments and the score of those comments

On reddit, for example, it seems impossible to get noticed once there are already a few highly-upvoted replies to the comment you're replying to.

Re: Hacking the Hivemind: Predicting Comment Karma on Internet Forums (2014) [pdf]

#33

Post by vivekd »

bcherny wrote: post #17
lacampbell wrote: post #15

I think karma mechanisms have had a net negative effect on online discussion. I don't want to have preconceived notion on how I should feel about a post - I want to make my own decisions. There's no escaping it now though, which is sad.

Given how much content is generated on online forums, it is natural to have a mechanism to surface better/relevant content (different forums define relevance differently). Otherwise you have to wade through a sea of memes and low quality content to get to something that is worth consuming.

True but the incentive system which gives people points for content tends to create an environment where people try and say what is popular or will get high karma instead of what is insightful and avoid what is unpopular or will get downvotes even when it is true and insightful.

Tying a system for surfacing high quality content to individual points seems like a recipe for disaster. I like facebooks system, where points and algorithms work together to try and surface content that users will like, but there is no actual running point system to try and game.

Re: Hacking the Hivemind: Predicting Comment Karma on Internet Forums (2014) [pdf]

#34

Post by lacampbell »

a1studmuffin wrote: post #27
lacampbell wrote: post #15

I think karma mechanisms have had a net negative effect on online discussion. I don't want to have preconceived notion on how I should feel about a post - I want to make my own decisions. There's no escaping it now though, which is sad.

I'd love to see an automated karma system that scored posts on correct grammar, spelling and readability. I guess that would discourage non-native speakers of a language from participating though.

My personal threshold for "how good should I be at a language before I start arguing online in it" is very, very low.

Re: Hacking the Hivemind: Predicting Comment Karma on Internet Forums (2014) [pdf]

#35

Post by 3131s »

mrkgnao wrote: post #29
tjalfi wrote: post #9

Earlier quoted context omitted.

This hasn't been true for me. My top 3 posts [0, 1, 2] are all fairly trite. All three of these were rated much higher than posts with extensive citations. [0] https://news.ycombinator.com/item?id=13538068 [1] https://news.ycombinator.com/item?id=13441744 [2] https://news.ycombinator.com/item?id=13594765 Edited: This post already has a higher score than https://news.ycombinator.com/item?id=14180432 which has five cit…

Compare this low-effort comment: https://news.ycombinator.com/item?id=14124411 to these: https://news.ycombinator.com/item?id=14149200 https://news.ycombinator.com/item?id=14016438 Getting "in" when a post has few comments and is going to attract a lot of comment readership quickly or stay on the frontpage for a while is probably more important than pure comment quality.

Absolutely true, and especially problematic because by their nature high-quality comments will take more time to formulate and therefore always be late to the party and get less visibility.

Re: Hacking the Hivemind: Predicting Comment Karma on Internet Forums (2014) [pdf]

#36

Post by slau »

a1studmuffin wrote: post #27
lacampbell wrote: post #15

I think karma mechanisms have had a net negative effect on online discussion. I don't want to have preconceived notion on how I should feel about a post - I want to make my own decisions. There's no escaping it now though, which is sad.

I'd love to see an automated karma system that scored posts on correct grammar, spelling and readability. I guess that would discourage non-native speakers of a language from participating though.

I don't think we should be overly PC. If someone doesn't understand the English language well enough to not write all of their thoughts in a single, unpunctuated wall of text, and doesn't rely on the red squiggly lines to indicate that something was misspelt... How exactly will they contribute to the discussion in a way that will be constructive and positive?

I'm not saying that their ideas aren't worthwhile, but simply that the medium won't allow for those ideas to be presented in an engaging manner.

There's a huge difference between the wall-of-text full of typos and misspellings (which seem to become the weapon of choice of native speakers as well), and the misuse of pronouns, or forgetting which word is an irregular preterit. I, for one, would be in favour of seeing the former ranked lower.

Re: Hacking the Hivemind: Predicting Comment Karma on Internet Forums (2014) [pdf]

#37

Post by fudged71 »

3131s wrote: post #32
Xcelerate wrote: post #11

Earlier quoted context omitted.

Haha no way. My highest rated posts have all been some vapid feel-good thing that I posted shortly after an article was submitted to HN. My detailed explanations that link to studies rarely get more than 10 points (same phenomenon occurs on Quora as well actually...)

I always wonder how much these factors affect a comment's score: * Proximity to the top of the page * Time of day posted * Number of sibling comments and the score of those comments On reddit, for example, it seems impossible to get noticed once there are already a few highly-upvoted replies to the comment you're replying to.

I've been kicking myself that I didn't take sociology in university to study conversational dynamics on threaded internet forums. It continues to fascinate me

Re: Hacking the Hivemind: Predicting Comment Karma on Internet Forums (2014) [pdf]

#38

Post by Sacho »

lacampbell wrote: post #22
stil wrote: post #20

Earlier quoted context omitted.

bcherny is saying that karma's primary function is content curation. sure, you might like a few downvoted submissions from time to time, but for most content karma scores will be saving you time sifting through a sea of content. removing content by hand requires an active and vigilant moderation team, user votes help mitigate that burden. Also, the power to 'delete' is most likely a slippery slope...

I see a lot of great content here greyed out. Including some of my own! At the end of the day, I'm not happy to follow the collective judgment of other users.

With your suggestion, you wouldn't be able to see the content at all. It would have been deleted.

Re: Hacking the Hivemind: Predicting Comment Karma on Internet Forums (2014) [pdf]

#39

Post by monk_e_boy »

minimaxir wrote: post #6
mrlyc wrote: post #3

That article reminds me of The Auditors in Terry Pratchett's "The Thief of Time" who try to analyse a painting by reducing it to its component molecules. There is no mention of tailoring comments and replies to the target audience. Once I adopted the mindset of a senior high school or college student and posted accordingly, I was able to increase my karma on Reddit from 1,000 per year to 1,000 per month. I still have…

I've written many, many articles/analysis about the karma systems of Reddit/Hacker News. Karma aggregates are meaningless internet points . Neither Reddit nor Hacker News, nor their users, give major impact to how much user a user has (aside for low karma aggregate thresholds for downvoting on HN), nor is there any correlation between the amount of karma and the quality of content that user outputs.

How about slashdot with 'insightful' / 'informative' / 'interesting' / 'funny', upvotes?

Does a system where the more karma you have the more power your upvotes have work?

Is there a better system?

Re: Hacking the Hivemind: Predicting Comment Karma on Internet Forums (2014) [pdf]

#40

Post by marklgr »

lacampbell wrote: post #22
stil wrote: post #20

Earlier quoted context omitted.

bcherny is saying that karma's primary function is content curation. sure, you might like a few downvoted submissions from time to time, but for most content karma scores will be saving you time sifting through a sea of content. removing content by hand requires an active and vigilant moderation team, user votes help mitigate that burden. Also, the power to 'delete' is most likely a slippery slope...

I see a lot of great content here greyed out. Including some of my own! At the end of the day, I'm not happy to follow the collective judgment of other users.

Rarely happens here, in my experience. Significantly downvoted content almost always has severe issues, either in form, tone, or quality. We sometimes might happen to agree with some controversial opinion that has been downvoted a good bit, but more often than not the tone was just not conducive to a good discussion. There's probably a few exceptions I guess, but overall the false positive rate is low enough for me.
Post reply on HN