FWIW, my current theory is that the problem is not the comments themselves, but the upvotes. There are often moderately stupid comments far down the page in any given thread, but they're comparatively harmless, like someone walking down the street mumbling to himself. The problem is that there's a type of stupid comment that attracts upvotes from a certain type of user. If we can recognize either the comments or the upvoters, we can solve the problem. I've been collecting a corpus of them for a while; it may soon be big enough to be useful.
I think the problem you describe is in essence the need for an automatic trust inference that recognises that worthwhile participants often certify badly and knows how to "prune" the set of tolerated participants down to those whose voting activity is high quality.
Advogato hasn't really been successful overall (Steven Rainwater suspended new account creation a year ago), but there are many successes in its history (it has been run with very little intervention for over 10 years, and generally succeeded in separating spammers from free-software types for most of that period) and its experience is probably worth studying.
It requires much more forum experience to identify what kind of comment causes insipid comment threads than it does to write a good comment in the first place.
That said, you could probably just measure someone's comment quality and their vote quality independently and get a lot of the benefit, with fewer errors from rigid categorization.
I see that you deem these comments to be comparatively harmless, but nevertheless to combat this one relatively simple idea is to implement a 'Request deletion' feature, sitting alongside the 'flag' link on comment headers, that would notify the user that the comment they've made isn't very useful/substantive/thoughtful, and if they delete it (as other users have requested) they can re-earn any karma points they may have lost (as an incentive to encourage them to go on with the comment deletion).
I think something like this would be very appropriate for a lot of occasions, it's something between a 'downvote' and 'flag' in intensity -- a downvote is too soft; 'flag' is too strong. I think it would send a good hint to have one reconsider what they've said, and have a lasting effect on the kinds of comments they make down the line.
> I've been collecting a corpus of them for a while; it may soon be big enough to be useful.
If you end up doing this I hope you only take into account the first few critical upvotes that send a particular bad comment to the top of the page. Because, I think 'upvoting' is a very passive, sometimes thoughtless action to a lot of people: if a comment is at the top, they'll read it first, and if it has non-zero substance to it (i.e. it attacks the article on a minor nitpick (which is your definition of a 'stupid comment' in this context, I think)) they may very well upvote it making the cycle complete.
I know I personally have dismissed down voting several "middlebrow dismissals" in the past because they seemed like valid counter arguments whereas outright useless comments like "lol this sucks" are easy for me to clearly distinguish. Maybe the rules just need adjusting to discourage mean comments whether or not the argument is valid? Is that the core problem?
Once I've decided not to read a subthread, it's a little frustrating to try to guess where to scroll to find the next constructive comment.
If collapsing can be linked to voting or thread positioning it'd be useful. Collapsing a thread means it's not very good[1] and thus should be further down the page.
There is an extension / addon somewhere that does thread collapsing on HN. A web search should find at least one.
[1] or maybe it just means that you're read it and are keeping the page tidy.
Additionally, if people mark as insightful things that are obviosly inane, that might help to hellban the voting rings, whereas hellbanning people just for agreeing with a particular kind of stupid comment might be less fair.
I've once or twice realized some longer term implication of something overnight after reading an article on HN; unfortunately, I don't remember at this point what the article was or what I realized in terms of the bigger implication. But in general, it seems like HN's structure makes it unlikely that you'll get a discussion if you comment on the article a day later. The solution may be to write a blog post and try to get it upvoted at the story level, but I'm not sure if that's the right answer to preserving the structure of which things are related to which others.
So what we're talking about would be a +1 on the "Affirming the consequent and denying the antecedent" anti-pattern, more or less?
Wikipedia has a nice list of fallacies which would make interesting tags.
One thing I know for certain is people like to imagine they personally invented their own fallacy, and get all outta whack if you identify their wisdom/insight as merely a poorly implemented manifestation of fallacy type #34 with aspects of type #11.
For example discussions of scalability always seem to devolve into steaming buckets of pragmatic fallacy and it would be interesting to filter out because its pretty tiresome. Someone who's a total noob might actually want it because a poor map is better than no map at all, or at least they can skim some definitions out of context. Or, personally I think amphiboly is hilarious and I'd filter it up by +100, although it adds absolutely nothing to a discussion other than joy in my life, however I can see how someone could get sick of it and want to filter it out if they were looking for actual knowledge rather than a rare laugh.
The further away a user is from an original deputy, the more X and Y increase before a new person is deputized.
Hopefully, this will give you a class of users who tend to vote intelligently and comment with information that adds to the converation. Votes by these users would carry more weight than the general community.
Edit: It also automates the process so you don't have to hand pick more than the original deputies.
But I think what you're essentially talking about is user reputation - the idea that the opinions of certain people weighs stronger than others. I think there's merit to that.
The upvotes of a person with a high reputation on HN would weigh more than someone else. Reputation could be determined by any number of ways (has authored comments of high value, is manually designated by HN as reputable, has been active in the last N number of days, etc).
The X in my equation accounts for that. You'd need several different deputies to upvote (endorse) a user before it'll deputize them. You couldn't deputize someone on your own.
The Y in the equation means that each deputy would have to have a history of upvoting someone more than once. So you couldn't immediately deputize someone.
There should also be a time limit. Only upvotes cast within Z hours would count in the system. Also, only 1 upvote counts within A hours. This prevents someone from going into a comment history and upvoting past comments.
Even though I think that I'm accounting for cheating, I also believe the system should be hidden from the users. This way no one knows if they've been deputized and they also don't know how to deputize someone else.
"Reputation could be determined by any number of ways (has authored comments of high value"
The trouble with a generic reputation system that values only votes is that it allows for people with any kind of comment (pedantic) to get reputation. My deputy system gives power to a few trusted users and allows that power to slowly trickle down. Hopefully, it'll sculpt the top comments and posts to what PG envisions for the site.
That said, I do think they often take away from detailed and serious discussion of the article, and there may be some problem with the way that they absorb upvotes and end up at the top of the page, but I don't see the general dynamic as particularly troublesome (unless you are trying to produce conformity in opinion among participants, the often problem of benevolent overlords who don't like to see strife amongst their minions).
It might be awkward to call someone out, but I'd love to see example(s).
My rate of upvoting things out of there that hit the front page is obscene, at least 50%. Assuming that is typical, that means very few people are monitoring that page. I personally would probably not do more of this for karma but I'd imagine many people would. And it may instantly solve the problem of hitting the frontpage being a lottery or determined by manipulation/voting rings etc.
It would help if there are official guidelines on up/down votes just as there are on submissions and comments.
[1] https://news.ycombinator.com/item?id=2365964 [2] https://news.ycombinator.com/item?id=1420725 [3] https://news.ycombinator.com/item?id=117171
I'd be curious on others' thoughts on schemes like this.
It's perhaps also worth noting that the "bogus upvote" problem is in some way related to some other social problems which we struggle with. Spam, and the wrong person using a password, both seem to me to be a reflection of social problems reflected by system design.
For spam, there's no way to ask the supposed origin mail server "is this from:, to:, message-id: combo a plausibly legit in-flight message?" Because some systems cannot handle that functionality efficiently, mail standards people have rejected it (and similar approaches, perhaps identifying a sequence of mail servers, perhaps involving multiple queries when mailing lists are involved) for the multitude of cases where it makes sense. But note that you do not need perfection here, to massively reduce spam volume. Note also that there are efficiency arguments, here, but the original protocol worked fine on machines with <64k memory and ~1MHz clocks.
Passwords and authentication protocols and buffer overflows are a more complex issue, and I'm drifting dangerously off subject already. But that's another issue, I think, where "design efficiencies" and "overly focussed thinking" sometimes bite us. (And perhaps sometimes benefit us, but how can we even measure this?)
Anyways, trying to get back on topic, personally, I have little faith in upvotes here, not even my own, so mostly I rely on friends to point out if there's anything worthwhile here, for me to read.
It's a triangle, someone clicked on it. Counters are a good first pass, for filtering bulk raw information and indicating things to focus on, but they are not the end-all and be-all for good ui design. Borrowing wholesale from stack overflow might be good, or maybe some variant would also be plausible (like, maybe: when replying, let people with parent posts weight child posts under theirs that they found helpful, giving some fraction of the total weight of upvotes in any child posts of their parent post to posts they want to highlight. This should be adjustable, in some simple way, or we live with accidents. There's also an issue when one person has multiple parents but maybe it's best to default to the top level eligible parent.)
My above parenthetical suggestion is not perfect, but it might help for distinguishing between some sincere and insincere replies. Another approach might be to give the user some "downvote" ability - abiding by the original site design, these need not be preserved for others (it's a personal decision, after all).
Ultimately, to really get this right, I would want to be able to segregate other voters into "agree" / "disagree" / "irrelevant" buckets based on a label which reflected my current point of view, and have the content displayed based on that categorization. But that's a much harder problem.
I think it's the same here. People can have opinions, but once says Microsoft is great and open source sucks or something like that, you're already past the point of no return, it's inevitable the thread or sub-thread will devolve into a flame war.
Certain threads are this way starting from the original topic. Like the threads, amidst the recent threads of falling programmer salaries (or perhaps mentions of the historically high unemployment rates) of the desire of the SV barons to raise the H1B cap. This again is guaranteed to degenerate into a flame war.
My unscientific study of HN is that someone can get a lot of karma for saying why they think Obama is good or bad or the like. A post on that guarantees a dozen upvotes. Someone making a cogent comment on how a company should limit the scope of what parts of C++ they use is not going to get as much karma.
I have been on BBSs since the 1980s. The real problem of a forum is this. It has a handful of very sharp, interesting people who like communicating with each other and who write good posts. Then you get some blabbermouths who post and post and post and fill the boards up with junk. Then the good people leave. The good people leaving is the problem.
I don't think flames are the only thing to be concerned with. Blabbermouths who have nothing to say kill boards as well. Have flamewars killed the Linux kernel mailing list? Just as long as things are kept under control. The NetBSD/OpenBSD flamewar got out of control.
Slashdot is a board that was good and had its founders leave and is all corporate and user-view baiting now. What are its topics right now? "Google's Nexus Q Successor Hits the FCC". Who cares? "John McAfee's Belize Home Burns To Ground". I can survive a few weeks without an update on McAfee. "NetBSD 6.1 Has Shipped". Something only the dwindling remainder of NetBSD fanbois care about.
Certain topics and upvoted threads, once existing, are going to turn into a flamewar no matter what. It has to be rooted out early on for anything to be effective.
(I see this as a sign that the flag link is properly used to mark dupes, spam, and inappropriate content, so there is a silver lining.)
As they say, you don't have to play the violin at all or as well as Yehudi Menuhin to be able to point out that he just missed a note. Criticism is easy, creation is hard.
I think it's important to note that comments can point out (even minor) factual inaccuracies in an article without being dismissive or insulting. I find that type of comment tremendously valuable for articles with topics outside of fields I am learned in.
Usually directly skipping half the page takes you to the actual discussion. I think it could be interesting to add a pivot that moves that comment and it's discussion to the bottom (but probably above downvoted comments) when it reaches a certain length. It could still keep it's votes and so on.
> With all these extremely knowledgeable people from different fields, I wish I found more big picture insights in the comments here.
The signal to noise ratio is still pretty good, and sometimes there are great posts.
Better flagging (and less upvoting) of the stupid politics / news dross would help.
With all these extremely knowledgeable people from different fields, I wish I found more big picture insights in the comments here.