Fixing Hacker News: A mathematical approach

http://gkosev.blogspot.com/2012/08/fixing-hacker-news-mathematical-approach.html

by spion • 14 years ago

155 102 14 years ago

gkosev.blogspot.com

PaulHoule• 14 years ago

This isn't too different from an algorithm I used to clean up mech. turk.

acabal• 14 years ago

Interesting, but with an important flaw the author acknowledges:

> The important downside of these results is that the people using the system were not aware that points are calculated in a different way.

As I mentioned in that HN-meta-whining thread, you can't engineer away jerks, and social problems need social solutions. People will naturally adapt to any sort of engineered restrictions and find away around them regardless. Personally I think the only way to maintain a very high standard of quality in a community is strict hands-on moderation and swift ejection of members who don't match the community's goals and vibe.

(In either case I think the thread this post refers to was about the negativity in comments, not the quality of submissions. Though I might be remembering wrong.)

spion3• 14 years ago

I think the comments voting system might also benefit from this.

But the downside of the testing results described is not necessarily negative. Maybe if people knew that their reputation and voting influence are on the line, they would try harder to write more constructive comments. Thats all speculation of course and the only way to find out is to experiment :)

anamax• 14 years ago

> Maybe if people knew that their reputation and voting influence are on the line, they would try harder to write more constructive comments. Thats all speculation of course and the only way to find out is to experiment :)

You could ask folks first. Not as reliable, but a lot less expensive.

Here's an unsolicited data point. I've got reasonable karma. I don't care about karma. And, I'm pretty sure that the some of the most interesting folks don't care either.

That's not quite true. I do care about all the karma whinging. I dislike it.

16s• 14 years ago

Perhaps we need a jerk flag. If enough people flag a user as a jerk for making inappropriate comments, then that user would be banned. People could upvote the offending comments all they like, but after X many jerk flags, that user is no longer a HN'er.

Edit: You could say that only experienced users (members for several years with X many points) may use the jerk flag. Or something like that.

debacle7• 14 years ago

I've never been on a site where the jerk flag wasn't abused.

If the majority of a site has $opinion_x, chances are people with $opinion_y will earn jerk points every time they express themselves.

andreasvc2• 14 years ago

That sounds rather like the current hellban system.

jorleif• 14 years ago

Why can't you engineer away jerks? I mean, we have democracy, free markets etc. and among their main contributions is to engineer away the effects of bad behavior. In a democracy, you won't get the best decision, but you will get an acceptable decision. In a free market competitors force business to serve the customer.

Similarly, a techical solution which gives jerks a smaller influence seems an excellent partial solution. Sure, you might still need policing, but it isn't either or.

debacle7• 14 years ago

Because anyone is free to sign up, and you have the tyranny of the majority.

And the majority of people are jerks.

cbs• 14 years ago

I agree totally with the limitations of a technical solution to a social problem.

Furthermore, even if the technical solution work to any extent, it doesn't work as well as it could if it were visible. The attitude of relying on monkey-see monkey-do for establishing your social boundaries is wishful thinking if your approach to unacceptable comments and people are to sweep them under the rug.

That kind of tinkering is bound to create a community where the only people talking are those who most adamantly adhere to the echo-chamber. We already know there are certain opinions on the periphery that we just don't discuss on HN, no matter how politely or clearly articulated they are. They go against the groupthink and are good way to get shadowbanned.

The lack of visible moderation has not served HN well. Community management isn't rocket science, but sometimes it does have to get a bit ugly. The price you pay for letting people disagree is disagreement. You can keep it within bounds, but wall-papering over the bounds every time they are crossed doesn't actually improve anything, it just makes the same wounds happen again and again.

Domenic_S• 14 years ago

Yep. Know who else tried to wave an algorithm at a community to fix it? Digg.

buro92• 14 years ago

"Personally I think the only way to maintain a very high standard of quality in a community is strict hands-on moderation and swift ejection of members who don't match the community's goals and vibe."

Many communities I've seen have done this in a way that drops-off over time.

Screw up any of your first 3 posts... you're out of here.

Posts 3 > 10... you're going to get a warning, and may still be out of here.

Posts 10 > 25... warnings only, only extreme cases get banned.

Above 25 posts you're largely left to it, though once in a blue moon someone will screw up so spectacularly that they'll get pulled up by the community.

Even then, the "pulled up by the community" is literally that. No moderators or banning, but the communities that enforce rules strictly for new members and than relax once you get comfortable within the constraints, generally are also the ones in which if you suddenly made an extremely racist post the solution is to be ripped to shreds by your peers rather than to censor and hide it... it reinforces the "that behaviour isn't acceptable here".

This is currently the most successful pattern I've tried. It works far better than other systems I've seen.

buro92• 14 years ago

Of_Prometheus: There is a deep irony in your post, in that it's marked as dead and not visible to many.

The question was, "How does that system deal with the old members > new members hierarchy that would develop? I think the problem with giving a member a longer stick the longer they stay/comment is that older members are given greater leniency when they do decide to be jerks, and new members who may very well develop into excellent members are kicked aside for possibly minor errors, thus creating an unfair advantage."

That hierarchy does exist. I'm not seeking to eradicate it.

The older members tend to be the ones that have established the tone and quality. This can be used to determine the rules that are used in the early stages of membership of a site.

But generally that hierarchy refers to inner circles and cliques who are protective of what the site evolves into for them, and this isn't addressed in any way by the system to integrate new members.

Old members will eventually leave, the question is more whether the value of the community is preserved when that happens.

paulsutter2• 14 years ago

People don't divide cleanly into good actors vs bad actors. Feedback mechanisms such as karma and downvoting on HN allows people to learn how to behave within the community. The better the feedback mechanism, the better people will behave.

Properly weighting the votes from each user would allow more effective moderation to be crowdsourced from the whole community. Everyone should be a moderator, but only with the degree of authority that he has earned.

That weighting can be done with a technical approach as has been proven by search* companies. How does Google deal with the fact that there is no moderation on the web in general, and a multibillion dollar incentive to manipulate their results? Google results are really clean and usable despite the above limitations. Their approach is purely technical.

This proposal is an effort to enhance the ability of the crowd to moderate HN more effectively by improved weighting.

*Note that Google is almost solely a ranking problem, and the the term "search" is a bit of a misnomer. If you could see the bottom ranked matches for any query I'm sure you would see plenty of unsavory content, but the first results (the ones you actually see) are really clean.

MayanAstronaut• 14 years ago

1. Google is not transparent with its ranking algo. We would need to be.

2. It actually is very cleaning divided, bots & humans. Bad humans have no real power due to point 3.

3. Google has a click traffic amount that is so high that it can only be taken advantage of by bots. Hackernews is tiny in comparison and any algo can be manipulated with just a few bad actors.

markkat• 14 years ago

I think this system could work.

One problem with community forums is that there is a drift to the 'majority taste'. As the population changes, so does the majority taste. This could put some drag on that drift, but you'd probably need to use historical voting data if it were to pull things back to the HN center. Otherwise it might just hold the community in a place where HN doesn't want to be.

brudgers• 14 years ago

I remain unconvinced that HN needs "fixing" though I am certain it could be improved. I also believe that the improvements are largely along the lines of better HTML (e.g. hiding comment scores) than more complex functions for calculating the value of a karma point.

Karma has the most influence on long term behavior - HNers have to learn community mores over time, and one vote one point is well suited for that task.

citricsquid• 14 years ago

This doesn't address the real issue. The issue isn't that people are voting up comments or submissions that don't fit with what HN is supposed to be, the problem is that those submissions and comments are being made.

Even if the top submissions and top comments fit squarely with what HN is supposed to be there is still going to be comments and submissions that don't and they will continue to be a problem. If a post has 50 comments and 25 are "bad" it's a problem, even if the 25 "bad" comments are at the bottom of the page because people will still reply to them and they will still be a part of the discussion.

spion3• 14 years ago

In a way it does address it: hopefully people will be more likely to think about what they're going to write if their reputation (and even more, their voting influence - something tangible) will be affected.

drharris2• 14 years ago

> The issue isn't that people are voting up comments or submissions that don't fit with what HN is supposed to be

What is it supposed to be?

JoelSutherland• 14 years ago

Interesting.

There are too many non-interesting, but pedantically correct comments that derail discussions. The community goal of an HN comment thread should be to unearth the most interesting, useful aspects of the submitted content. Correct but non-interesting, non-contributing comments should be down-voted.

rachelbythebay• 14 years ago

Isn't this the same pattern as yesterday's "bozo" post? The general principle there was "B players hire C players and so on". Granted, we're talking about posters and topics instead of "players", but couldn't it be the same thing in a nutshell?

"People are voting up comments or submissions that don't fit..." sounds just like "People are hiring people who don't fit..." to me.

Or am I wrong?

cousin_it• 14 years ago

It seems to me that trying to solve HN's problem by voting would be more similar to solving a company's bozo problem by internal policies, like performance reviews. If bozos can overrun any internal policy of the company, and the only solution for companies is not hiring bozos in the first place, then the only solution for HN is not allowing bozos to register accounts. (Not sure how you'd achieve that though.) That fits well with my impression that HN's early awesomeness was due to an initial pool of awesome people attracted by PG's writings, not due to voting.

debacle7• 14 years ago

Can you link to the post from yesterday? I didn't see it, but it sounds interesting.

adeelk2• 14 years ago

> The issue isn't that people are voting up comments or submissions that don't fit with what HN is supposed to be, the problem is that those submissions and comments are being made.

If people stop voting up such comments, people will eventually stop making them.

ScottWhigham2• 14 years ago

> If people stop voting up such comments, people will eventually stop making them.

That's like saying, "If we start putting people in jail for using illegal drugs, people will eventually stop using illegal drugs." Unfortunately that's just not how it works - some people comment (or use drugs, etc) because they can't help themselves, rather than for karma/recognition/trolling/attention. Idealism is a good place to start but but, long term, you have to recognize it as such.

huggyface2• 14 years ago

even if the 25 "bad" comments are at the bottom of the page because people will still reply to them and they will still be a part of the discussion.

Here's a good example-

http://news.ycombinator.com/item?id=4407733

andreasvc2• 14 years ago

That's rich, because it rather looked like you were the one trolling in that exchange.

debacle7• 14 years ago

I didn't feel like that was bad discussion, per se. You were getting ganged up on for having a counter opinion (one that would have been accepted at face value on HN a year ago) and responded poorly, but overall good points were made.

TheComedian• 14 years ago

You seem to assume that the pattern of top submissions and top comments is unrelated to what users post. It seems to me that as quality of the top submissions and top comments improves, the quality of the worst submissions and comments will improve as well. When there is no longer a reward for posting "bad" material behavior will change accordingly.

cabalamat• 14 years ago

> If a post has 50 comments and 25 are "bad" it's a problem, even if the 25 "bad" comments are at the bottom of the page because people will still reply to them and they will still be a part of the discussion.

What is a "bad" comment anyway? Your idea of a bad one is bound to be different from mine, and pg will no doubt have a third view. Would it be possible to present to each user those comments that that user is likely to want to see, based on that user's voting behaviour?

randomdata2• 14 years ago

What if users with a registration date around your own or prior had a more significant weight with respect to their votes? That should, at least in theory, keep the content fairly consistent with the quality you came to know around the time you registered.

recoiledsnake• 14 years ago

>This doesn't address the real issue. The issue isn't that people are voting up comments or submissions that don't fit with what HN is supposed to be, the problem is that those submissions and comments are being made.

The solution then is to autohide comments with low scores(configurable on the page) like Slashdot does. Some of the HN comment threads have grown too long these days. It would be nice to skim the comments that the community thinks are the most valuable.

vannevar5• 14 years ago

Then you'd be punishing all newbies, not just the bad ones. Instead, hide low scoring comments from low-ranking users, but show them to high-ranking users so that they have the chance to upvote them.

vannevar5• 14 years ago

Even if the top submissions and top comments fit squarely with what HN is supposed to be there is still going to be comments and submissions that don't and they will continue to be a problem.

Only if those comments are made public to everyone. If this system were adopted, it would be sensible to also implement a filter so that low-scoring comments only appear to high-ranking users, who are presumably responsible enough to filter the bad ones out and raise the good ones to visibility. I'd be happy to see fewer comments per post if the quality increased.

shell0x• 14 years ago

I'm not sure if it would be a good idea to switch the algo behind HN.It worked well for several years,but the spectrum of users changed. I'm not against a new solution, but you will run into this issues again and again, because it are social problems, not technical ones. Just my 2 cents.

stefanve• 14 years ago

I always liked the idea of adding meta data to your up vote (funny, insightful etc) just choose from a list, and people can filter out category's they have no interest in. together with a system where you have to earn voting rights (to become a 'citizen' of a site), for instance: you have to be a member for at least a year and you start with a veined number of up/down votes.

parfe• 14 years ago

Slashdot solved that problem over a decade ago. Semi-Random users receive moderation powers and can assign a post to several categories (Flame-bait, Troll, Funny, Redundant, Insightful, Informative). Any other user has the power to review moderator choices. Get enough negative feedback and you lose the chance to be chosen as moderator in the future.

You also cannot moderate and participate in the same conversation. Posting a comment revokes any moderation in the thread.

You can change your user preferences to augment the moderation choices, such as -5 to funny and +3 to insightful.

CmdrTaco provides history on how the system evolved: http://slashdot.org/moderation.shtml (1999)

snogglethorpe• 14 years ago

The can't-mod/post-simultaneously rule is super annoying though, and I wish it were at least a bit more subtle. For instance, maybe something like "can't mod the same sub-thread you've posted in" or whatever (to stop people from modding down their opponent in an argument).

kzahel• 14 years ago

I think it makes more sense to just update the posting and commenting guidelines to more accurately reflect what kind of posts and comments are "desired" by the old guard. Not all of us have been around silicon valley for 10 years and understand the unwritten etiquette of HN!

thornofmight• 14 years ago

It's so upsetting being a 21yo just getting started in the industry. I finally find this awesome site with like-minded, motivated, and successful people, but all anybody can talk about is fixing it!

It seems like I'm always on the tail-end of communities. Is this just simply a case of the grass is greener for you veterans of HN?

CodeCube• 14 years ago

remember, it's a cycle. I'm 10 years older than you, and when I was 21, newsgroups were where it was at. Then I found Slashdot ... these exact discussions happened back then. Those communities spawned the diggs, reddits, stackoverflows and hackernews(es?) of today.

Some new community is brewing somewhere as we speak, just participate wherever you still enjoy while you still enjoy it. And when you find a new community that you enjoy, simply participate ... in ten years, I guarantee that you'll make a comment exactly like this to some young gun ;)

debacle7• 14 years ago

It happens. If you're a risk taker and willing to be a part of multiple social networks at once, you can be on the bleeding edge of good communities but how much are you really losing out on? The next good community will spring up sooner rather than later, and we'll all visit it, for a while.

Communities seem to have a 2-5 year lead in cycle, followed by 1-3 years of good, healthy community. Then they get mainstream, bloated, and meta.

But they never really die, which is the saddest part. I can go to social networks I signed up for over a decade ago, and people are still stomping there.

And they are angry. Angry at being deserted, or angry because they put all their eggs into one basket, but either way you wind up with little hovels all over the Internet that are remnants of the gold rush for good discussion.

Some day, the Internet as a whole will find its El Dorado. One thing's for sure - it's probably not Facebook.

steve8918• 14 years ago

I'm double your age. It's fairly inaccurate to say that all everyone can talk about it fixing it. Sure, some people gripe about it, and some aspects of the community might have shifted over time, but you'll still find the highest quality comments of any online community out there.

spion3• 14 years ago

I'm sorry, I made the headline a bit sensationalist to attract attention. Paul Graham said he was looking into various options for the voting system, so I thought I'd pitch in and suggest my idea. That is all really - I love hacker news. No need to be discouraged or upset.

mbesto• 14 years ago

People have been talking about "fixing it" since day one. It's a reoccurring theme and typically goes in waves.

However, I've been on here for 3+ years. I've seen the community decline unfortunately. It's only been in the last year that I've seen more than 100 comments (although this probably a cognitive bias) consistently frontpage articles and in my opinion is a bad thing. The quantity has certainly increased, but technically the quality hasn't diminished, it's just much harder to dig through the crap to find it.

TL;DR - The signal to noise ratio has gotten much worse, but the signal quality seems to be the same.

lmm• 14 years ago

Part of me thinks that communities have a lifecycle. I feel like I caught a few on the way up - 4chan and facebook - and some on the decline - K5, plastic, fark. Only /. seems to go on and on like it always was.

Sadly HN is feeling more and more like it's in its twilight years, if only because of all these endless posts on how to "fix" it. Particularly since I'm seeing people suggest a $5 signup charge. That killed K5 stone dead.

debacle7• 14 years ago

You can't fix the HN problem without people.

If pg really wants to preserve the feel of HN, he needs to develop a system that analyzes the behavior of users to determine their acceptance level of other users. This will create a network effect, where the most influential users will drive the site.

He then needs to seed the system with 5-10 users who he considers paragons of HN (I can think of five off the top of my head, but his are almost certainly different), and tweak as necessary.

lysol• 14 years ago

Another approach is to limit submissions via some manner. Metafilter still retains a similar level of content as it did when it started around 12 years ago, and they accomplished this by a) Requiring signups to pay $5 b) Maintaining strong, public moderation where deletion reasons are clearly shown and frequently discussed ad nauseum on a section of the site clearly designated for talking about the site itself.

It may not be everyone's cup of tea but the site has grown leaps and bounds and hasn't yet strayed from its original goal, which is albeit more open-ended than HN. I'm not suggesting users pay $5, but there are more nuanced approaches than trying to squash the problem with the math hammer.

politician• 14 years ago

One problem is that there isn't a feedback loop describing why a comment or article is bad, just that it is bad. Negative context-less signals lead to superstition which induces drift away from the values of the Creators.

One way to provide context could be to allow people to tag their downvotes with an adjective like "irrelevant", "mean", "notreddit", etc. Then, critically, only the author of the comment would be able to see this additional feedback.

zerostar073• 14 years ago

I 've seen many solutions, and personally i believe it's a psychological matter of attention overload: too many submissions / comments -> less time spent on critical review -> hasty "cat picture" upvotes.

Let's create artificial scarcity: let each user be able to vote only on a small random sample of submissions/comments

Other ideas: Require approval by 2 blindly selected users before showing, or allow multiple dimensions in voting[1]; but these don't seem to work.

Anyway, this is a deep issue, essentially political [2]. It's also fun to consider the dizzying election procedure that the Venetians used [3].

[1] http://textchannels.com/

[2] http://en.wikipedia.org/wiki/Criticism_of_democracy

[3] http://en.wikipedia.org/wiki/Doge_of_Venice#Selection_of_the...

ansible2• 14 years ago

...let each user be able to vote only on a small random sample of submissions/comments

Slashdot first pioneered that. In case you are not familiar with the scheme, all logged-in users are occasionally asked to vote with their 5 allotted points. Users with high karma are given a chance to vote more often. The points expire in a day or so if they are not used. They also have a meta-moderation system, where you judge how others have used their votes.

They've had the present system for a while (I want to say over 6 years, but I don't frequent the site as much these days). The overall quality of comments there is... OK. Not great, but not as bad as it was for a while around the turn of the millenium.

zerostar073• 14 years ago

actually, it shouldn't be a per day quota because heavy users will be disadvantaged. The idea is to have a steady rate of voting for as long as a user is on the site.

icebraining• 14 years ago

The quality of Slashdot comments depends a lot on your default threshold. If you hide every post with score < 2, it's decent. If you include lower scores, it drops considerably. -1 is almost masochism.

debacle7• 14 years ago

The level of snarky comments there in contrast with actual commentary is too high, though. If a story has 25 comments chances are one is modded insightful though it is factually incorrect, two are modded as funny, and the rest are hidden.

I remember when real scientists used to visit /. Now, not so much.

finnw• 14 years ago

It's not really a random sample, because you can choose which 5 posts to use your votes on.

zobzu• 14 years ago

/install slashdot ;-)

oh yeah, i might get down votes - but - its still one of the massive comment systems that more or less works. more.. or less. :)

stcredzero2• 14 years ago

Unfortunately, the most famous installation of this code is saddled by the editors of Slashdot.

russtrpkovski• 14 years ago

We need to access to user activity data in order to determine root cause(s)and come up with potential recommendations. All proposed solutions without any empirical evidence is based on speculation.

Kilimanjaro• 14 years ago

Here is a simpler formula:

    vote = max(10,floor(karma/1000))

If I get an upvote from tptacek, I'd get 10 pts but a down vote will send me to oblivion at once.

At the same time noobs don't have voting power until they get enough karma, increasing comment quality.

Trolls who get below -100 will be automatically hellbanned.

zerostar073• 14 years ago

Via that strategy old users become tyrants; even if we assume they are wise, there is the danger that karma whores will become the next generation of tyrants.

peterwwillis• 14 years ago

Moderators.

Moderators, moderators, moderators, moderators, moderators, moderators, moderators, moderators. Moderators, moderators, moderators, moderators, moderators!

eranation• 14 years ago

Newcomer question, how can I know if my comment / post fits the HN "good" or "bad" qualification? is it something describable? or is a bozo doomed to be a bozo?

SkyMarshal• 14 years ago

Everything is learnable. For the poster-child of good posts, see Patio11's comments:

http://news.ycombinator.com/threads?id=patio11

Also, know the official etiquette/guidelines:

http://ycombinator.com/newsguidelines.html

think-large• 14 years ago

I would be very interested to see this implemented in a side by side comparison. I think it would be great to see the type of content that would float to the top of HN.

I also have to say that I love this site. I feel that the content of the posts have made me a better programmer in general and I really appreciate everyone's input. Even if the discussion threads spiral out of control quickly.

It would be nice to have expandable comments so that when people begin name calling we have an easy way of filtering them out on screen and just viewing the people who directly address the post.

tkahn6• 14 years ago

There is no way for the community to effectively exert influence on the behavior of new members. That is the problem.

A complicated algorithm or formula will not fix this.

New members have the ability to upvote. If there are enough of them with interests and standards contrary to the current culture of HN, they will upvote each other and bootstrap their own culture. Eventually they and the people they subsequently attract to HN will overwhelm the original culture.

This problem will not be fixed unless there is a threshold system for upvoting or PG and the Y Combinator mods take a more aggressive approach in removing articles.

vannevar5• 14 years ago

There is no way for the community to effectively exert influence on the behavior of new members. That is the problem. A complicated algorithm or formula will not fix this.

Sure it can. If you hide low-scoring comments from low-ranking users, but show them to high-ranking users, the high-ranking users can exert influence on the low raking-users by upvoting quality comments and downvoting poor ones. It becomes a self-organizing system with a bias towards quality, rather than the open loop 'majority rules' system currently in place.

Tichy• 14 years ago

Just had an idea: fixing parts of HN by allowing ads. But not for products, the ads would be for discussions in other communities. The idea being to lure trolls and other unwanted members away into other fishing grounds.

tokenadult• 14 years ago

The Hacker News guidelines

http://ycombinator.com/newsguidelines.html

have been updated pretty recently, as I see one significant change in the guideline text that responds to controversies that came up over the last two months. The Hacker News welcome message

http://ycombinator.com/newswelcome.html

gives an overview of the community experiment here, summarizing the site guidelines.

http://ycombinator.com/newsguidelines.html

The Hacker News FAQ

http://ycombinator.com/newsfaq.html

gives some additional details about how Hacker News is administered. The welcome message distills the basic rules into a simple statement: "Essentially there are two rules here: don't post or upvote crap links, and don't be rude or dumb in comment threads."

In recent discussion,

http://news.ycombinator.com/item?id=4397542

several HN participants proposed further revision of the guidelines, and also proposed making the guidelines more visible during the article submission and comment submission process. For example, have the "add comment" form field prominently display a link to the guidelines, perhaps with a snippet of text referring to the guidelines most related to comments, and similarly have the submit form

http://news.ycombinator.com/submit

prominently display a link to the site guidelines and a brief description of what kind of submissions are most desired.

Weighted voting fixes based on user behavior signals do seem like a good idea (that is what pg is thinking about), and we might as well discuss how to make those fixes mathematically correct. The technical tweaks can best be reinforced by ongoing efforts at user education, including possible revision of the guidelines, and, in the opinion of several HN users, simply making the guidelines more visible to everyone who submits an article or posts a comment.

Edit to reply to first comment kindly made to my comment:

What kind of comment is most often upvoted, and by whom, is an empirical question. I don't have access to the data to resolve the question of what kind of comment is most readily upvoted. The bestcomments page here on HN

http://news.ycombinator.com/bestcomments

shows a current snapshot of what the current community, based on current communication of the guidelines and current technical features of the HN software, has upvoted the most in the most recent several days.

transpostmeta• 14 years ago

The problem with votes being weighted upon previous upvotes of content submitted by that user is as follows: The chance of upvoting a post is related to its length and complexity. One-off puns or silly remarks are much more readily upvoted than long, insightful but demanding comments on complex matters.

ljf• 14 years ago

Well the difficulty is really that we don't know that any more (since we can't see comment vote counts). I'd say it's not inverse to your suggestions - some 'higher' HN posters do seem to get away with 'comedy' comments, but many newer members who don't get the 'HN tone' get downvoted into oblivion.

I think since we can't see the vote numbers people are less likely to upvote longer comments now - but I don't imagine them losing out to jokes.

I'd still think that over all, good comments are going to attract more votes that joke comments, and with less risk of being downvoted for 'making this like reddit'.

paulsutter2• 14 years ago

Should a person's vote be weighted based on how well they comment, or how well they vote? Either could be argued, both might be tried.

Making a dataset available, even to a limited set of folks under NDA, could allow some interesting experimentation.

robomartin• 14 years ago

HN isn't perfect, but it isn't that bad. I've seen countless fora degenerate over the years. This one seems to be in check for now.

I see the voting system as one of the problems. More accurately, the obvious emotional or fanboi voting trends along some topics. Blind down-voting, as I see it, is a problem. If you down-vote you ought to be required to provide an explanation, otherwise, don't. Topics such as many subjects surrounding Apple draw in people who will mercilessly down-vote well reasoned comments simply because they happen to be anti-Apple. The same applies to the occasional political discussion in which anything that isn't aligned with liberal thinking is castigated.

The biggest problem I see with this is that it does not promote intelligent contrasting posts. Those with different opinions either leave (or stop posting) or fall-in with the crowd and choose to become part of the collective. If one was after collecting points and accolades all you have to do on HN is be all over anything that relates to Apple and talk like a Liberal. That's the formula. Oh, yes, you also have to be 100% pro FOSS.

The fact that it is so easy to define is kind of sad. Then again, maybe that's what the YC guys wanted HN to be. If that's the case, there are no issues. If the idea behind HN was to bring together pro-Apple, pro-FOSS, politically Liberal hackers, then there's nothing to fix.

If, on the other hand, if HN is intended to promote diversity of thought along a wide range of subjects of interest to hackers and entrepreneurs, the voting system needs to reflect this.

Someone• 14 years ago

"If you down-vote you ought to be required to provide an explanation"

Here is an idea:

- require downvoters to provide that explanation.

- Use some simple automated method (e.g. run the text typed through a language recognizer, and see whether it thinks it is English) to prevent simple key-banging as a response.

- by default, do not show the explanations, but do show a 'has downvotes' indicator and a 'show downvotes' control (could be a single UI element)

- when showing downvotes to a user, do not show author names.

- allow users to flag downvotes as inappropriate.

- let the powers that be manually check flagged downvotes (hopefully, there aren't many), and take appropriate action against either the downvoter or the flagger, or both.

robertskmiles• 14 years ago

I really think that this could work. It reminds me of a "Web of Trust" system, where you have a few known trustworthy individuals, and the reputation propagates out from them.

rossjudson• 14 years ago

Seems to me the missing element from your analysis is simple: Users don't need to have exactly the same experience. Picture a graph where each user is a node. When I freshly arrive at the site, it tells me to pick out a few users to place in my "circle", to use Google terminology. I do this by finding users who write comments I think are good, or by looking at a list of some of the most popular users. I can then set the degrees of Kevin-Bacon controller to set expand my "circles" outward by joining in people that others in my circle trust, multiplied by a reduction factor (farther out means less influence).

Only votes by members of my extended circles count towards my "score" for a given item, weighted by distance. If something is upvoted in my perspective, I can see who and why, and break that connection (remove a person from my circles).

Finally, I can publish a group of users I trust and give it a name, which makes it easy for others to follow well-known groups.

I think it would work. A user has to put in a little effort, to find people they think have good judgement. It's very easy to knock out bad judgments by breaking links.

If I were less tired I'd think this through. The downside to all of this is processing time, of course.

dinkumthinkum• 14 years ago

Maybe it would help if we decided what the problems really were? Everyone seems to agree that it is completely broken; I'm not seeing it. I see a few, what I consider to be, very bad comments occasionally and they tend to be down-voted out of visibility, so it seems like the current system sort of works? Is the problem more related to there being people going against the accepted groupthink? It feels a little like "grass is greener" to me. No community is perfect; to me it seems like the main issue, and I don't think it's a real problem, is the there seems to be a growing amount of non-technical people, which is OK but it will tend to change the content of discussions quite a bit if the trend continues. Overall, I'm obviously in the minority, I don't think there is a dramatic problem.

The post that started this recent wave, someone was, I suppose, genuinely upset about not receiving good comments to a "look at my project post." However, a user looked through posts and actually found a lot of praise ... So, I don't know what to think about that.

jhuckestein• 14 years ago

I think posting this on YC's Demo Day is unfortunate timing. I'm pretty sure this won't be noticed by anyone at YC.

6ren• 14 years ago

Similar to PageRank.

jedharris• 14 years ago

Lots of interesting and potentially useful points. But everything depends on knowing or controlling the quality of HN: Is it deteriorating? How do we increase it? Etc.

So: We need to be able to measure the quality of HN. This doesn't sound any easier -- maybe it isn't. But the process of figuring out how to measure the quality will take us a long way toward telling us how to maintain and increase it.

Without measurement, we don't really have a basis for agreement on whether a change to the mechanism is good or bad.

drblast• 14 years ago

I'd posit that threaded discussions are a problem.

Particularly with the nerd community, the ability to reply and pick apart what the parent post said almost ensures that every discussion devolves into inanity.

I try to make it a point, if I'm going to comment, to just leave a top-level comment and not respond to replies. It works well enough, and it might be worth considering as a rule.

sesqu• 14 years ago

Beyond the problem of not counting votes as contributions (the knights of /new should be rewarded), I feel this scheme has a major downside in using the Wilson lower bound: namely, it rewards quick comments.

It's well known that the number of votes a submission or comment gets is very time-sensitive, and the Wilson bounds are substantially affected by that number. This leads to little-seen content gaining their author very little influence, which may be purposeful but probably isn't.

A further little-mentioned complication is the existence of thresholded voting: some people distribute votes freely, while others manage their vote histories by limiting themselves to voting on exceptional content, and yet others use contextual voting, touching only misrated content. Given the dependence on voting, I feel this aspect has not been sufficiently investigated.

efsavage• 14 years ago

I'd like to see a system where your vote is simply part of an estimate as the ultimate quality of a comment. If things Bob votes on tend to also gather other votes, Bob's vote should be seen as an indicator that other people will be voting for that item soon. So his vote basically counts for more early on, but as others join his vote effectively counts less as the actually tally reaches the estimated target.

You could take this to the level where if Bob and Mary both vote for something, it's a sure hit, but either one on their own means little. Some kind of Bayes implementation.

Fixing Hacker News: A mathematical approach

Related Stories