Hacker News new | past | comments | ask | show | jobs | submit login

Another subtle feature addition: a flame-war detector. Graham has been consistently deploying and updating proprietary software that determines whether there is a flame war, where people argue heatedly. When these flame wars take place (which Graham says can often get ugly and personal), the story in which the commenting is taking place is moved further down the page.

Remember the above the next time someone claims that a rogue contingent must be flagging stories on a particular topic because it is further down the page than other stories submitted at a similar time and with a similar number of points.




It might just be me, but this seems like the kind of feature you wouldn't want automated. How do you algorithmically tell the difference between a flamewar and a spirited debate?


It would go a long way to partly explaining why Apple stories get buried so quickly while ones related to Google get to the top and stay there. For whatever reasons Apple engenders strong emotions, positive and negative, while Google does not, at least to that degree.


Did you see the stuff related to Google demanding the Windows Phone Youtube app be removed? Massive threads, accusations of shilling all over, along with claims of organized flagging. At least 2 of the major contributors to those threads/flame wars (recoiledsnake and CloudNine) have been hellbanned, and the whole thing was rather unpleasant to read. It also spilled over into most other Google posts on HN, which was quite a few as it was in the middle of Google IO. Along with that there is the inevitable comment on every story about a Google service suggesting that Google will shut it down in 6 months. There is a sizable subset of HN which isn't crazy about Google.


His solution was to put flamewars further down the comment list, but if almost all of the comments get flagged as flame war, then it's free to flourish.

Maybe what HN needs, could be some way of giving people who want the heated discussion, a way to express these seperate from rest. Maybe a tab that shows controversial comments, so you can switch between the "calm" and the "heated" discussions.


There are other forums for flamewars, HN doesn't need to be another.

Most of these problems aren't a big deal on a case by case basis because, really, what's the harm of one more google vs microsoft flamewar in the scheme of things. But they're a problem in the example they set to other users, especially newer users who are still malleable. They'll write better comments if they think that this is a forum for good comments. At least that's the philosophy.


> Maybe what HN needs, could be some way of giving people who want the heated discussion, a way to express these seperate from rest.

I like this idea. I'd say all who are hellbanned should see other hellbanned users' and they can nuke each other from the orbit in their own little sandbox.


You can set "showdead" in your profile to see hellbanned users. But you still can't reply to them (that's kind of the point).


Paul Graham wrote an essay in 2002 advocating the use of Bayesian classification for detecting spam (http://www.paulgraham.com/spam.html, http://www.paulgraham.com/better.html). I bet he's doing something similar to detect flame wars, but with more modern techniques and more CPU power.


Given how little attention Paul Graham seems to give to HN's tech, I very much doubt there is anything sophisticated there. Apparently HN is a single-threaded Arc process with in-memory state [1] and flat files, running on a single server.

It's so low-tech that HN runs a log tailer that bans your IP if it sees a "suspicious" number of requests, the threshold being something like 4-5 reqs within 1-2 seconds. Personally I don't do anything weird (I just browse in Chrome on my Mac and on my iPad), but still I manage to get my IP banned frequently. Instead of fixing the algorithm, Graham added a way to unban your IP [2].

A hypothesis was put forward recently that Graham has intentionally designed HN to be technologically backward to attract serious hackers and repel everyone else. In that light, everything -- the "unknown or expired URL" errors, the impossible-to-hit-without-zooming-on-a-mobile-device up/down arrows, the IP banning, the awful UI, etc. -- makes perfect sense. I am not convinced that it's all intentional, though.

[1] https://news.ycombinator.com/item?id=3623268

[2] https://news.ycombinator.com/item?id=4761102


My IP has been banned on occasion because I tend to read HN by opening a bunch of tabs. When my browser inevitably becomes unresponsive and has to be restarted, that rush of a couple of dozen new requests as the tabs all come back online triggers the ban. Makes me think twice when presented with Chrome's "restore the pages you had open" button.

It's funny how the stability of modern computers and operating systems has allowed my reading and research habits to evolve toward the representation of state in the form of open browser windows and the groupings of tabs therein.


HN is the most ruthlessly prioritized software project I've seen. It wouldn't surprise me at all if pg thinks anti-flamewar Bayesian classification is more important than fixing the URL expirations.


> the impossible-to-hit-without-zooming-on-a-mobile-device up/down arrows,

One unfortunate effect of these is that I tend to ignore the single downvotes, and rationalise those away as accidental clicks.

Really, I need to train myself out of that. When I get a downvote I should re-read my post to see if I've been clear, and to see if I said what I meant, and if I still mean it.


I have downvoted by accident only once in my time as a HN user, and it was on an iPad.

These days, out of fear of clicking the wrong arrow, and because zooming in and out is a chore, I up/downvote sparingly when I am on a mobile device.


> Personally I don't do anything weird (I just browse in Chrome on my Mac and on my iPad), but still I manage to get my IP banned frequently

Me too, with s/Chrome/Opera/ and such. I avoided it by setting the maximum number of connections to news.ycombinator.com to 1 rather than the usual 128 in Opera, which seems to avoid this problem for me.


If there is a good reason for the rapidly-expiring links, I'd be interested to know. For me, it's the #1 biggest annoyance.


Flamewars likely have a mix of upvotes and downvotes that is very different than a spirited and friendly debate. For example, there might be two general groups of commenters, with most of voters upvoting one half and downvoting the other half.


Be careful with that one though: You'll get the same behavior when you have a group of people trying to engage in a serious discussion which is being trolled by a single poster with multiple accounts/IP addrs. In that case you would want to somehow distinguish one group from the other rather than penalizing both of them.


"Your mother was a hamster and your father smelt of elderberries"


One non-NLP route might be to track the rates of votes on the posts in question. Flamewars probably elicit more downvotes than normal against all the comments in that subtree.


Bayesian text classification?


If a flamewar in the comments can be detected, why not demote the comment thread, rather than (or in addition to[1]) the article?

[1]In addition to because presumably the logic is that the article content is flame-bait.


SWAG: demoting the entire thread is the best way to snuff out the flame war. My personal observation is that part of the motivation for trolling/flame wars is attention seeking ... i.e. you don't have a flame war without an audience.




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: