Hacker News Reader

Comment by minimaxir | Hacker News Reader

Upvote manipulation doesn't work on Hacker News because both the voting ring detector is strong, and flagging is significantly more powerful and serves as counteracting force to upvotes and serves as a check to submissions which legit aren't good for the site. (both of these are in contrast with Product Hunt, which turns a blind eye to voting manipulation and asking-friends-for-upvotes-but-not-explicitly is often recommended as a promotion strategy on that site: https://medium.com/startup-grind/my-startup-launch-on-produc...)

There is also a Hacker News "growth hacking" tactic of linking to /newest on Facebook/Twitter instead of linking to the submission directly to avoid the voting ring detector, which also doesn't work and is blatantly obvious: http://i.imgur.com/08pAFOw.jpg

View on HN

> Upvote manipulation doesn't work on Hacker News

I'm pretty sure that's unequivocally false. It's difficult enough to make it work that it probably wouldn't be worth it, but I'm close to 100% sure I could force something to the top if I wanted to. I don't particularly want to spend the time automating accounts to make them "look natural" and I don't really want to shell out the money to spend on IPs.

But I know for a fact the same thing is happening on Reddit with great success every day. I could spell out how, but I don't think the HN mods would be too keen on that. Any hacker should be able to figure it out, though.

Furthermore, if you produce something solid, all you'd really need is the first <10 upvotes to get on the front page and let it catch on "organically." I can promise you that is happening _all the time_ on HN.

I'm glad you keep your Twitter search running and keep calling people out that blatantly ask for upvotes; the easiest way to kill spam is just to make it difficult, but to say that manipulation doesn't work is just not true.

> I just don't have the time or the money to shell out and buy IPs.

You need a Hacker News account to upvote, and obviously having a large number of new accounts upvote something would trip a voting ring detector.

> Furthermore, if you produce something solid, all you'd really need is the first <10 upvotes to get on the front page and let it catch on "organically."

Much easier said than done. The median score of a HN submission is 1-2 points last time I checked. (the discoverability of HN submissions on /newest is an issue, but that's another topic)

And if it's actually solid, it's not the market for vote brigading.

I agree manipulation does happen, but it doesn't mean it works and instead silently penalizes the submission to death.

> You need a Hacker News account to upvote, and obviously having a large number of new accounts upvote something would trip a voting ring detector.

Correct, but what if I gave myself a couple or few months to prepare accounts? Do a search on blackhatworld (or another spam forum) for "aged accounts" and you'll see that you can generally buy them in tiers - the older they are the more expensive they are.

I'm not saying it's worth it or I think it's something people should do, but saying it's not possible or doesn't work is absurd.

I think it'd be fairly obvious that something suspicious was going on if a large number of dormant "aged accounts" just happened to login for the first time in months to upvote a specific thread.

Except this is why there is an economy of aged accounts....

It directly affects the cost of the promotional service

I'm pretty sure upvotes from different users are unequally weighted when it comes to getting a story onto the front page, and that upvotes from fresh users aren't counted with nearly the same weight as someone who's been here a while and has a track record of predicting which stories will perform well on the front page. There've been times that I've given a story languishing on /newest its 4th upvote and it pops onto the front page, and there've also been times that I've seen stories get 10+ upvotes in short succession and get stuck on page 3. It also wouldn't surprise me if receiving a lot upvotes from newly-created accounts, or accounts that have all upvoted the same things, ends up tripping the voting-ring detector.

If you look at how PG originally created the algorithm in Arc this behavior makes a lot of sense, especially when alongside it is the explanation that HN will cancel out votes that look shady. (I could be wrong, but I believe this is the case).

Basically the algorithm looked like:

Score = (p-1) / (t+2)^g

p = points (the -1 is to negate the point of the original submitter)

t = time

g = gravity (or some gravitational constant)

So the p value matters a lot; especially when there's a smaller t (a new story). This gives newer stories a chance to bubble up with fewer points against the ones that have 1,000 points but have been on the front page for days. It's a pretty ingenious algorithm, actually.

It makes sense, given that formula, that even if all p values are weighted equally, moving p from 3 to 4 would make a big difference, and would be able to push something new to the front page. You'll see them now and then with 2-3 upvotes on the front page, because they're very new and the front page is relatively stale.

Anyway, I'm ~100% sure spammers can get around the vote ring detection, but I don't think that's happening on HN in a major way.

These were cases where the submission was both older and the point value was lower than other submissions that were not on the front page, though.

It's possible that flags, voting ring detector, or spamkiller were negatively impacting the other story, but it definitely isn't a straight votes & time formula anymore.

> I'm pretty sure that's unequivocally false. [...] I could spell out how

What makes you sure? If you know something we don't, we need to hear it—the community relies on us to get this right. It's also personal: I've poured countless hours into combating vote manipulation, and if you know my code isn't working, you bet I want you to spell out how.

The upside of a public discussion outweighs the downside because (1) if you tell us things we don't know, we can start investigating for them—it's what we don't know that's worrisome; and (2) community members might respond with valuable knowledge of their own. That's also why we haven't buried the OP (which, in case users are wondering, we do know about and have lots of data on).

I'll email you.

For what it's worth, I don't think HN is in any immediate danger, nor do I know of any active, large-scale vote manipulation happening currently. But to say something like "vote manipulation on HN does not work" is a very bold claim.

Ok, I look forward to it. Edit: still waiting!

I understand that HN's ranking algorithm isn't 100% disclosed because people post startups here and this forum needs to be a fair environment. I say this to make clear that I do get the security ramifications of revealing too much.

But it sounds like you have a bit of understanding when it comes to this topic, which I'm very interested in understanding more about generically speaking. I've always been curious how vote manipulation detection _actually works_, and the kinds of things that can be done to combat vote hacking of various kinds.

I say this as someone vaguely considering the idea of making a small HN-like forum at some point (still figuring out the why and how).

I'm sure where there is a will, there is a way...

>Upvote manipulation doesn't work on Hacker News because both the voting ring detector is strong, and flagging is significantly more powerful and serves as counteracting force to upvotes and serves as a check to submissions which legit aren't good for the site.

You say this, and it may or may not be true, but of course, Hacker News isn't releasing the code to prove it.

It would be a lot easier to get around vote manipulation detection if the code was public.

Crypto industry is a step ahead in this game, they would release white paper proving that even the creators of the system can't manipulate upvotes.

This isn't crypto, and there's no concrete definition of manipulation that would support proof (unlike the kind of threats crypto deals with); it's a fuzzy problem domain.

If that's true, it implies that vote manipulation detection is already easy to get around. It's just relying on security through obscurity and probably direct human intervention to cover for unpublished flaws.

Aren't you holding us to an impossible standard here? The problems we're talking about are not cryptographic; they're much closer to, say, law enforcement. Has anybody figured out how to model such things with mathematical guarantees?

It's just depressing that a site that calls itself Hacker News, and that centers around discussions about software, web applications and security, by so many intelligent and capable people, is less open and less hacker friendly than Wordpress.

And I understand why - the problem is that votes mean something to startups that post here, they have a potential real money value. But it's unfortunate that any insight into how the HN staff deals with the problem of being gamed is denied to the community at large, out of fear of being gamed even more. Plenty of people here would be willing to learn Arc, audit and contribute to the codebase if you let them, and that scrutiny could be beneficial, not just harmful.

The idealism here is really noteworthy, and (as just another commentator/user here) I must say I like it very much.

As you yourself note, however, I can't see any possible way to make what you're describing practically work out: there is absolutely NO way to tell the good guys from the "growth hackers". :(

The other problem is that the bunch of people willing to do this work - which (I say this non-critically, please note!!) is similar to IRL work for a community organization or similar, in that it's not attractive or "widely scoped shiny impact that looks awesome on paper" - is much, much smaller than the bunch of people with evil grins just waiting for more tidbits to use against everyone else.

Security by obscurity is not obscurity. It would be a lot easier to detect and fix flaws in the vote manipulation detection if the code was public.

Reply on news.ycombinator.com