Ask HN/PG: What causes this sudden drop-off behaviour on HN submissions?

by dazbradbury • 14 years ago

93 55 14 years ago

dazbradbury6• 14 years ago

It seems unlikely that this is algorithmic, hence, I would guess if a post is deemed to not be meeting HN standards / guidelines, moderators can "demote" a post, essentially removing it from the front page. So in this case, was the title simply too link-baity? And who judges that?

This is obviously a guess, and looking through the FAQ (Official and unofficial!), I can't seem to see it stated anywhere.

Another possibility could be user flagging, but that would seem too easy to game maliciously.

I think whilst some elements of your filtering you want to keep secret to stop spammers, it would be helpful to understand this throttling behaviour to help people submit better content. Transparency is often helpful to create less work for both sides, so if an answer could be given, I think people would be interested to know!

andrewcooke• 14 years ago

i don't know how current it is, but i believe that hn source is included in the arc download - see http://news.ycombinator.com/item?id=1390685 - so you could try looking there to see how flagging is handled.

rpm43213• 14 years ago

My impression is that quite a bit of the source for flagging, vote-ring, and spam detection isn't included in the arc download, for obvious reasons.

dazbradbury6• 14 years ago

Good point - could be in here, but it's 1:30am and I probably shouldn't be learning Arc right now! Will take a look in the morning.

    (= flag-threshold* 30 flag-kill-threshold* 7 many-flags* 1)

    ; Un-flagging something doesn't unkill it, if it's now no longer
    ; over flag-kill-threshold.  Ok, since arbitrary threshold anyway.

    (def flaglink (i user whence)
      (when (and user
                 (isnt user i!by)
                 (or (admin user) (> (karma user) flag-threshold*)))
        (pr bar*)
        (w/rlink (do (togglemem user i!flags)
                     (when (and (~mem 'nokill i!keys)
                                (len> i!flags flag-kill-threshold*)
                                (~find admin:!2 i!vote))
                       (kill i 'flags))
                     whence)
          (pr "@(if (mem user i!flags) 'un)flag"))
        (when (and (admin user) (len> i!flags many-flags*))
          (pr bar* (plural (len i!flags) "flag") " ")
          (w/rlink (do (togglemem 'nokill i!keys)
                       (save-item i)
                       whence)
            (pr (if (mem 'nokill i!keys) "un-notice" "noted"))))))

weaksauce• 14 years ago

Probably the algorithm, check out these(old) posts on it...

http://amix.dk/blog/post/19574

http://news.ycombinator.com/item?id=1781013

dazbradbury6• 14 years ago

Those posts describe a smooth drop off as a function of time and votes - in some instances (such as the one linked to) that clearly isn't the case.

legierski• 14 years ago

I'm wondering if there's any description of the voting ring detector used by HN

_sentient• 14 years ago

Are there any YC companies competing directly in the space? 42Floors comes to mind, although I believe they are focused exclusively on commercial real estate at the moment.

If there is a YC company that competes directly with you, I could see incentive for a mod to suppress your post. </tinfoil hat speculation>

dazbradbury6• 14 years ago

There have been several YC-backed rental/property startups, but it seems a little far-fetched that pg/yc would get admins to suppress competitor stories.

I'm sure there are plenty of examples of competitor start-ups reaching, and staying on, the front page that could rule that theory out.

ximeng2• 14 years ago

It's not tinfoil hat speculation. Sharp dropoffs also happened when the story of an AirBNB house being robbed broke. Rather than just seeing one story dropping off we need a way to see every time this happens to make our own minds up over whether it's above board.

rpm43213• 14 years ago

From everything I know about PG and YC, I highly doubt they themselves would take that kind of action unless a post was outright slander. It just seems Nixonian.

That said, there is obviously quite a bit of goodwill on HN for PG and YC. You could call it fanboyism if you were feeling uncharitable. To me, that seems a much more likely scenario - an anti-YC submission getting a bunch of flag votes, rather than an intervention by admins.

petercooper• 14 years ago

Possibly, though it's not necessarily a mod trying to "suppress" a story. There's probably a social mechanism at play.

When posts on HN are flagged, their ranking gets hit a lot. And a reasonable share of people on HN are YC affiliated. Part of the value of YC is a sense of camaraderie and I suspect YC participants are more likely to support/flag posts that are supportive/dismissive of YC related enterprises.

It's a bit like if someone posts a big troll post about Ruby. I, and other Rubyists, are more likely to flag it. People who buy into the whole "Ruby community is evil" nonsense are more likely to vote it up.

wavesounds2• 14 years ago

But then why is this post still on the front page?

mrmaddog• 14 years ago

It looks like this "Show HN" post wouldn't have a high "gag" score [1], so either the post has been flagged multiple times (unlikely), or, most likely, the majority of those early votes came from the same location, causing the voting ring safeguards [2] to kick in.

[1] http://news.ycombinator.com/item?id=1782058 [2] http://news.ycombinator.com/item?id=3894695

dazbradbury6• 14 years ago

Unfortunately, I know this wasn't the case - I didn't provide the link to anyone in our office, which has it's own IP, and I know they weren't aware of the HN post until afterwards. I also wouldn't publicly use this as an example if I knew I had been rigging the votes!

So maybe it was just a case of HNers flagging the story as it became popular, as the title was fairly link-baity? I was under the impression that flagging a story kills it rather than suppresses it, but regardless, if that's the case then it makes sense.

Edit: One comment in your example link by ColinWright seems to state 2 or more "flags" can cause this behaviour: http://news.ycombinator.com/item?id=3893207

If that's the case, it's probably best people don't know about it, as used maliciously it would be fairly disruptive.

jtheory• 14 years ago

> Unfortunately, I know this wasn't the case

You can't know -- just because it wasn't in your office doesn't mean it didn't happen.

One of the first readers to see it may have mass emailed his/her colleagues who all read/upvoted when they got into work in the morning, for example....

pg• 14 years ago

You set off the voting ring detector.

dazbradbury6• 14 years ago

Thanks for responding, and I don't mean to waste your time, but that seems like a bug. We have a single ip, that 4 people can use, and none of them knew of the hn post until after the post was demoted. In fact, I don't think any of them have a hn account!

On top of this, I didn't share the link with anyone, and the votes appeared to be totally organic.

Our target market is not the hn crowd, and this subsequent post was merely out of curiosity, so I have very little to gain by lying.

Edit: if you or your team wanted any more information for debugging purposes I would be happy to oblige, just let me know if I can help, though I understand if this is fairly low priority!

jQueryIsAwesome2• 14 years ago

It maybe set off the vote ring detector because your costumers seem to be located all around London. So it may be possible that some of them shared the submission with other London users or that because they know your company they up-vote it; making it look like it was ringed but it was just selection bias.

Anyway this is a tricky problem; because at the end of the day intent is a fundamental part of vote rings and that can be really hard to see by algorithms alone.

flipside• 14 years ago

Can you share anything about how you define a voting ring?

zitterbewegung• 14 years ago

Probably doing so might tip off the people who try to game HN so I don't think its a good idea to share that kind of information.

shanelja8• 14 years ago

There was really no need to respond so pointedly or awkwardly, the parent posted a genuine question and it really begged a slightly more polite response.

I understand that you are an extremely busy man and I don't mean to bag on the owner of the site, but at the risk of getting hell-banned or massively down voted your response came across as being unhelpful and rude, if it had been someone else making the comment they would likely have been down voted for it, or at the very least treat with indignation and ignored.

Perhaps you didn't meant to come across this way and perhaps I'm in completely the wrong mindset about it, but such a short answer does nearly nothing to help the parent, you simply stated they had set off a trigger and gave no details as to why, what they could do in the future to prevent this or any kind of statistics or advice to aid them.

dfc• 14 years ago

If pg gave "details as to why, what they could do in the future to prevent this or any kind of statistics or advice" he would be giving advice on how to get around the voting ring detection. I imagine he probably does not want to help people evade the process he worked to put in place to cut down on spammy/fraudulent submissions.

spitx2• 14 years ago

    "but at the risk of getting hell-banned or massively 
    down voted your response came across as being unhelpful
    and rude, if it had been someone else making the
    comment they would likely have been down voted for it,
    or at the very least treat with indignation and
    ignored"

What? When did it come to this?

Has HN become a Reddit-style circle-jerk, increasingly elbowing out all dissenting opinions even when they are sensibly argued and corroborated with plentiful facts?

When did karma war chests become purses for silencing or pariah-izing, differing or even polarizing thoughts?

I have always thought you had to reason your way to the proverbial top of the heap of consensus-seeking opinions.

Not carpet-bomb your way to it.

Legitimate big wig or a green horn. Same credo.

In fact "thought leaders" should invite criticism and discourage herdism.

This is a disquieting trend.

If you can't rationally present your case or think a curt one-liner will do since your tech peerage will add some pedigree to an otherwise lousy and hollow argument, you should go elsewhere.

This should be a default understanding among all civil discussion online.

Unfortunately we live in a hyper-partisan era where discord is detrimental. A prevailing view that directly contrasts yours is a dent in your sense of self-worth.

HN should strive to be exemplary in this regard.

xnxn• 14 years ago

> I'm in completely the wrong mindset about it

My friend, I think this is probably the case ... you've mistaken terseness for rudeness.

JohnHaugeland• 14 years ago

Terse responses are pretty common among programmers, and are not generally interpreted as disrespectful.

apeace2• 14 years ago

An inconvenient truth of HN seems to be that 95% of voter ring detections are legit, while the other 5% are not. This is understandable. As jQueryIsAwesome said, it's a tricky problem.

Another inconvenient truth is, the demigods of HN often see the 95% as the 100%. No matter what defenses fellow HN users can come up with, pg likely gave his response in the way he did because he assumed something about you. Knowing pg's writings, he may have apologized or at least explained the problem if he had identified one. To say that terseness is pg's typical mode is...wrong.

I'm glad you pointed this out. pg has done a lot for this community, so we can't criticize too lightly. But it is a shame that in this instance, he didn't take that extra step for a fellow hacker.

jacquesm2• 14 years ago

He gave an answer when he was under no obligation to do so.

Anybody that is monitoring the effect of their submissions and is asking what went wrong is asking about potentially dangerous information. HN is a pretty fine balancing act, it would not take much to wreck it and the secret sauce serves us all well.

richardw• 14 years ago

I respectfully disagree. He pointed out the issue. Had I received that response I'd now know what to watch out for. He's not a support guy who is committed to helping all-comers to realise their HN-posting potential.

Try that with Google when you fall foul of their mystical ruleset. "you're not in the index and your business doesn't exist anymore. bye." And that's when you pay them - HN you don't pay and it's just one post, not your whole business model.

OzzyOsbourne• 14 years ago

I don't understand this concept? Anybody care to clarify?

edit: specifically, What constitutes a voting ring? Why would HN want to find and prune them?

martincmartin• 14 years ago

You try to unfairly increase the upvotes on your submission, by getting people who don't read HN to register and upvote your submission. Or, you create a bunch of HN accounts, and log into each one and upvote your submission.

Getting on the front page of HN can drive a lot of traffic to a site, which can turn into money, so there's a financial incentive to game the system.

TallboyOne• 14 years ago

What I want to know is how I get tons of HN traffic daily months after my post is off the front page.

evilduck• 14 years ago

Search engine results. The comments are indexed and are often niche enough to dominate certain search terms.

MaysonL• 14 years ago

Probably it's search generated...

supersaiyan• 14 years ago

Can I ask what js framework/proj. you are using to graph the data?

jpalacios• 14 years ago

looks like highcharts JS

JeremyMorgan• 14 years ago

Because you're on the 15th page by then?

zerostar07• 14 years ago

You 're all entrepreneurs (or wannapreneurs). Why doesn't one of you create a service that will employ mechanical turks as seed to bootstrap a new website by using it and providing honest feedback? I know i would pay for that. Being on the frontpage in HN will not guarantee seed userbase; it's more of an ego boost.

Ask HN/PG: What causes this sudden drop-off behaviour on HN submissions?

Related Stories