Comment by glangdale | Hacker News Reader

Comment by glangdale | Hacker News Reader

Parent Full thread

I wish someone would create a branch of mathematics related to the study of relationships between a bunch of "nodes" and "edges".

Jokes aside - and I'm aware that the FB folks have plenty of good graph theorists - how hard can it be to spot these kind of non-organic actors in the social graph? If I'm paranoid, I would guess that there's probably little distinction between FB's valued corporate customers and advertisers and these 'information operations' in terms of their presence in the social graph.

In any case, it seems like you could do a lot of worthy stuff to distinguish between content that spreads relatively slowly among people who appear to post things judiciously (and might even comment about it in ways that add content) vs various types of bullshit or low-quality content. Again, paranoia suggests that weeding out reflexively transmitted garbage and 'fake news' (for whatever value of 'fake' you're holding today) cuts a bit too close to the business model...

> how hard can it be to spot these kind of non-organic actors in the social graph?

Very hard. Graph theory contains some of the hardest problems in computer science. For example, maybe you want to find the largest group of nodes that are all connected to each other. We don't know how to do this. If the largest clique has n nodes, our best approximation methods can find log(n) of them, which is pretty useless.[1]

Spam detection at FB scale is difficult because with so many users many actual accounts will appear unusual in some way, and of course all data can be faked. Google and Yahoo have worked on this for years, yet spammers can still evade their systems: [2]

> there's probably little distinction between FB's valued corporate customers and advertisers and these 'information operations'

The distinction is pretty clear. FB knows exactly who its advertisers are since they pay FB money to promote their posts. It's true that fake news increases engagement, but FB is tackling an easier problem--identifying active operations making millions of fake accounts.

[1] https://en.wikipedia.org/wiki/Clique_problem

[2] https://news.ycombinator.com/item?id=13802227

I agree with you in principle, but I think the problem is more one of scale than of complexity theory.

Sure the clique problem is hard to approximate in arbitrary graphs, but the graphs that appear in social networks are far from arbitrary. And indeed, in such scale free networks the clique problem becomes quasi polynomial or polynomial (http://link.springer.com/chapter/10.1007/978-3-642-35261-4_6...). Approximate solutions are also much easier to find.

On the other hand, the sheer size of Facebook's social graph means that anything other than streaming algorithms or local search is pretty much intractable. That makes solving these kinds of problems into interesting engineering challenges.

Good points. Pedantic aside: scale-free is a dangerous assumption for two reasons. First, there may be fairly large deviations from power law in real data, and any such deviation renders the problem intractable. Second, botnets can modify local and possibly global properties in adversarial ways, so as to hide in or make large areas that are not scale-free.

Actually, the graph theorists at FB solved the problem in a few days, but various government agencies insisted that the code should not unearth anything potentially negative about their operations. Its now been years, and they are still trying to get it to work within that constraint.

how hard can it be to spot these kind of non-organic actors in the social graph?

It's easy to find suspected bad-actors, but very hard to get it accurate enough to be useful.

The problem is that organic actors are extremely diverse in their behavior, and non-organic actors are usually "human driven" and actively trying to avoid detection.

Graph-based features are useful but to get a useful system you need to combine graph features with other things (temporal behavior, NLP, image analysis etc).

(I work in this area)

It's easy to detect and delete these accounts...just as easy as it is to create new ones!

It's a cat and mouse game...and sometimes it makes more sense to watch what the mouse is doing vs trying to catch it :)

"...just as easy as it is to create new ones!"

I don't do Facebook, never have. But a few months ago I wanted to add login via Facebook to an app I was developing so I signed up for an account in order to get access to the developer stuff. I verified my email address, entered a mobile number, and my account was immediately blocked with a message that I had to send in copies of government id and such. Nope.

My question is, if it's so easy to open new fake accounts why did a legit account opening like mine get blocked?

It isn't. Facebook blocked all my legit accounts (those I used for ads and groups) and also expects a ID copy now.

Not long ago, I wrote about my own frustrations with Facebook, and their request for an ID:

"Facebook activated my dormant account and it won’t let me deactivate it"

http://www.smashcompany.com/philosophy/facebook-activated-my...

https://news.ycombinator.com/item?id=13965134

> My question is, if it's so easy to open new fake accounts why did a legit account opening like mine get blocked?

It's simple really, facepook already collected the so called shadow dossier on you, now they need to attach it to your real identity so that it could be sold to advertisers.

Is it a smart way of telling us that is quite difficult to detect and delete those accounts ?

I mean creating a facebook account is getting ridiculously difficult, several hoops to jump in a row and when you manage to do so it hits you with a brick and instablock the account before you even log in asking to send government issued identification with a picture.

The problem is always false positives.

I'm currently in Las Vegas, and I was struck once again with the realization about how big the place is. The airport is huge, the streets are huge, the buildings are huge. And, it's all air conditioned.

Power here is cheap, but getting from point A to point B takes time.

If we could figure out a way for an individual to "waste" time on getting from point A authentication to point B authentication, then we'd have a way to prevent these types of issues from occurring. Or, make them heat up and have to stop to cool down, as it were.

As it stands, identity on the Internet is completely broken, and the idea of a federated Internet is just a twinkle in Google and Jeff Bezos' eyes.

Reply on news.ycombinator.com