3 ms·

You can test how well this would work by collecting a corpus of whatever sort of comments you think HN could do without, and then checking to see how many of th

by pg 14y ago

You can test how well this would work by collecting a corpus of whatever sort of comments you think HN could do without, and then checking to see how many of those commenters would have run out of karma using whatever combination of initial karma and cost to post you have in mind. I'd be open to experimenting with charging karma to comment if someone did a study of that type and showed that it would work. (I realize it's not a perfect test, because people would make different comments if they knew comments cost karma. But that is a desirable direction to have error in.)

timtadh 14y ago
I think it would be good to do a targeted posting rate eg. 30 posts per hour[1] or something of that nature. If last 20 min when regressed predict under 30 posts in the next hour it should cost 0 karma to post. However, for every post/hour over the target rate should cost the user some karma eg. karma_loss = floor(alpha*max(PREDICTED_RATE - TARGET_RATE, 0)) where alpha is a tune-able "cost" constant. The posting page should of course warn the user of the karma loss and confirm that they really want to post. [1] 30 posts/hour stands for whatever target rate you want EDIT: it occurs to me you could apply a learning algorithm to learn what alpha should be based on experience of what yields the closest posting rate to the target rate. you could also reward people for posting when it gets slow.
- jerf 14y ago
  Don't forget second-order effects. Your proposal encourages people to post quickly, to try to get in before the costs start kicking in. Encouraging people to crank posts out faster is probably not a net win. I've noticed that trying to engineer communities isn't entirely unlike trying to secure software. You're not defending against straight-line attacks, you're defending against a hostile adversary with more time, resources, and collective brainpower than you, who will actively react to any move you make. The miracle is that any community ever works at all/any software is ever secured at all.
wonginator1221 14y ago
I think that there's also a chance that charging karma for a comment could also alter the quality of discussion here on HN. Although it may reduce comment "spam", commenters would have a stronger incentive to follow popular opinion to ensure that the cost of posting a comment is recouped. This could potentially reduce the number of dissenting (yet still intellectually provoking) comments. This hivemind mentality is a major problem in many online communities, but I do think that the hiding a karma on is definitely a step in the right direction.
- dwwoelfel 14y ago
  Users also might try to nest their comments under one of the top comments instead of under the submission where they belong.
- pbhjpbhj 14y ago
  IMO one is trying to do with karma what Google are doing with SERP positions - reward valuable content. If you do what they've done and personalise results then gaming the system becomes a lot harder; the only real way to win has to become producing quality content. Metrics such as rank, average, sd, account age, up & downvotes could be combined by the reader in order to personalise story listing and comment rank/display on the story page. That way if I want to reward a persons comments with a karma boost for average comment karma I can do so, but you could just do upvotes or just rank on inverse of downvotes or whatever. This way the site users generate competing algorithms, algorithms will be adjusted and 'evolve' to generate the prefered ranking of comments. Those who dont want to write one can pick a recommended or default algo. If user written algos are too expensive to use (though I'm thinking it would be client side in js) then a selection of algos could be offered by the site.
thaumaturgy 14y ago
I was going to modify my HN natural-language search engine to try to build this dataset for you, but then http://news.ycombinator.com/item?id=4491894 http://news.ycombinator.com/item?id=4491894 just happened (the submission from uncrunched.com where Mike Arrington weighs in on the whole stupid Google Ventures tempest-in-a-teapot). A bunch of those comments are adding absolutely nothing of value to HN. A bunch of them are just picking on Arrington as a controversial figure. And, of those, most of them are by accounts with close to 100 karma, or less. For what it's worth, I've thought for quite a while that forcing users to spend karma might be a worthwhile experiment, mostly because of comment threads like that one. One of the big failures over at Reddit is the huge number of people who can create an account and post a completely vacuous comment in a matter of seconds; on a popular thread, there can be thousands of comments voted to 0 or less, just crowding out any good content that might be left. On the other hand, karma-as-currency potentially encourages further gaming of the site; it could conceivably lead to e.g. fewer unpopular-but-not-wrong comments, and more popular-but-not-insightful comments.

by pg 14y ago

timtadh 14y ago

I think it would be good to do a targeted posting rate eg. 30 posts per hour[1] or something of that nature. If last 20 min when regressed predict under 30 posts in the next hour it should cost 0 karma to post. However, for every post/hour over the target rate should cost the user some karma eg. karma_loss = floor(alpha*max(PREDICTED_RATE - TARGET_RATE, 0)) where alpha is a tune-able "cost" constant. The posting page should of course warn the user of the karma loss and confirm that they really want to post. [1] 30 posts/hour stands for whatever target rate you want EDIT: it occurs to me you could apply a learning algorithm to learn what alpha should be based on experience of what yields the closest posting rate to the target rate. you could also reward people for posting when it gets slow.

jerf 14y ago
Don't forget second-order effects. Your proposal encourages people to post quickly, to try to get in before the costs start kicking in. Encouraging people to crank posts out faster is probably not a net win. I've noticed that trying to engineer communities isn't entirely unlike trying to secure software. You're not defending against straight-line attacks, you're defending against a hostile adversary with more time, resources, and collective brainpower than you, who will actively react to any move you make. The miracle is that any community ever works at all/any software is ever secured at all.

wonginator1221 14y ago

I think that there's also a chance that charging karma for a comment could also alter the quality of discussion here on HN. Although it may reduce comment "spam", commenters would have a stronger incentive to follow popular opinion to ensure that the cost of posting a comment is recouped. This could potentially reduce the number of dissenting (yet still intellectually provoking) comments. This hivemind mentality is a major problem in many online communities, but I do think that the hiding a karma on is definitely a step in the right direction.

dwwoelfel 14y ago
Users also might try to nest their comments under one of the top comments instead of under the submission where they belong.
pbhjpbhj 14y ago
IMO one is trying to do with karma what Google are doing with SERP positions - reward valuable content. If you do what they've done and personalise results then gaming the system becomes a lot harder; the only real way to win has to become producing quality content. Metrics such as rank, average, sd, account age, up & downvotes could be combined by the reader in order to personalise story listing and comment rank/display on the story page. That way if I want to reward a persons comments with a karma boost for average comment karma I can do so, but you could just do upvotes or just rank on inverse of downvotes or whatever. This way the site users generate competing algorithms, algorithms will be adjusted and 'evolve' to generate the prefered ranking of comments. Those who dont want to write one can pick a recommended or default algo. If user written algos are too expensive to use (though I'm thinking it would be client side in js) then a selection of algos could be offered by the site.

thaumaturgy 14y ago

I was going to modify my HN natural-language search engine to try to build this dataset for you, but then http://news.ycombinator.com/item?id=4491894 http://news.ycombinator.com/item?id=4491894 just happened (the submission from uncrunched.com where Mike Arrington weighs in on the whole stupid Google Ventures tempest-in-a-teapot). A bunch of those comments are adding absolutely nothing of value to HN. A bunch of them are just picking on Arrington as a controversial figure. And, of those, most of them are by accounts with close to 100 karma, or less. For what it's worth, I've thought for quite a while that forcing users to spend karma might be a worthwhile experiment, mostly because of comment threads like that one. One of the big failures over at Reddit is the huge number of people who can create an account and post a completely vacuous comment in a matter of seconds; on a popular thread, there can be thousands of comments voted to 0 or less, just crowding out any good content that might be left. On the other hand, karma-as-currency potentially encourages further gaming of the site; it could conceivably lead to e.g. fewer unpopular-but-not-wrong comments, and more popular-but-not-insightful comments.