Comment by blauditore
4 years ago
I'm really curious about voting ring detection. Oftentimes I've seen posts from the same company hitting the front page over and over again, and none of them was particularly interesting, nor was the company any of HN's "love children" such as Stripe. I can't recall any specific example, but when looking a bit closer it was usually some small- to medium-sized startup with maybe 10-100 people, which would technically be enough to make a big dent into a post's score.
On the one hand I'd love to know more about the detection algorithm, on the other hand that information would be inevitably abused by those shitposters to game it.
You can email dang and ask that he looks into a specific case in more detail (although usually the answer is "looks natural")
"I'm really curious about voting ring detection."
I am curious about this as well.
When I read "voting ring" I think of accounts trading votes - a group of accounts votes for each others' stories.
But your description (possibly accurate) is that asking people to vote for a story is a "voting ring".
Once in a while my company makes an announcement of some kind and emails that announcement to some portion of our customer base. Sometimes that announcement includes a link to an HN story that I create with an invitation for discussion. We never make any mention of voting or scoring.
This makes good sense because we don't have a forum and everyone prefers the HN user interface anyway.
Is this a voting ring ?
If this is not a voting ring, per se, is it likely this still trips the voting ring detection ?
Is this poor HN etiquette ?
Ring detection works for people who don't know it exists. And thus works well a lot of times. But is very easy to bypass, and I won't tell you how.
And if I recall well, dang has already said several times that he will not disclose the algorithm https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...
> But is very easy to bypass, and I won't tell you how.
That really depends on how it works, and whether you know how it works. Simple signals such as IP location and temporal distribution of votes are easy to use for detection, but also easy to manipulate. On the other hand, if you employ a graph of user associations based on votes and comments in the past, you can detect clusters among them, voting collaboratively. This is way harder to circumvent in the long run, but also extremely difficult to implement accurately.
It's pretty easy to generate new accounts here