Hacker News Reader

Comment by sigmar | Hacker News Reader

How about the SEC opens an investigation into whether Twitter is defrauding investors/advertisers by including bot accounts in user metrics? Or is that too hard to prove?

View on HN

Facebook has been caught dead to rights a few times, like this one:

"According to a recently published report, Facebook says they reach 1.5 million Swedes between the ages of 15 and 24. The problem here is that Sweden only has 1.2 million of ’em"

"Facebook’s Ads Manager says that the website is capable of reaching 41 million Americans between the ages of 18 and 24. The problem is there are only 31 million Americans of that age." [1]

Nothing ever happens, other than maybe some grumbling from paying advertisers.

[1]https://mumbrella.com.au/will-facebook-ever-stop-bullshittin...

>Why should a person's mere nonexistence prevent you from serving an ad to her? You have built a scalable technology platform; it hardly seems fair that your growth should be limited by the human species's comparative lack of scalability.

(Via Matt Levine, tongue firmly in cheek)

http://www.thevab.com/wp-content/uploads/2017/09/Facebooks-R...

I didn't read your source. But I'm sure that for those age gaps the number of accounts could be explained by underage kids lying about their age to be able to create an account.

a lot of people create alternate personas for varying reasons. There are a sizable cohort of people that like to create accounts for online fantasy. also, there are many game accounts (some that even produce money)

trying to police them off would be a mistake by facebook.

>a lot of people create alternate personas for varying reasons.

I'm sure they do, but I doubt 30% of them do. That's what you'd need to balloon from 30 million to 40 million. The 30% also assumes that EVERY 18-24 year old American has a Facebook account in the first place.

You'd be surprised. A lot of people are technically inept.

I've seen people create new accounts every time they lose their password. I also had a friend who'd create a new account every time she had a personality change of some sort (new hair? new profile!). Another friend owns a couple businesses and has accounts for each one (Pages are completely lost on him...). My aunt has entered her birthday incorrectly so she's supposedly 19 years old right now. A neighbor has a "family account", which is really just the wife posting pictures of kids.

Don't take "technically inept" to mean "my abilities, just a bit lesser". Take it to mean "no ability to reason about software applications whatsoever".

If they aren't logging into those accounts any more, then Facebook's advertising isn't "reaching" them. That explanation doesn't paint Facebook in a good light either, because presumably they are measuring how many impressions they serve and to who, right?

I agree, it's not good on Facebook to advertise inactive accounts (maybe they don't?), but it's not really in their immediate interest to do so.

facebook makes around 80% of its revenue from mobile

Start by assuming virtually every one below 21 that has an FB account has at least a couple of accounts - One with their public persona and one their parents won't be allowed to see. Add in accounts pet owners have for their pets etc. Then add in the bot farms. You'd get there pretty fast.

Another way to slice those data - 33% don't have accounts, 33% (9.9 million) have just one account, and 33% have the remaining 30.1 million accounts spread between them - on average, 3.04 accounts per person in that group, but one or two people/groups may have thousands of accounts.

Edits: Correcting math. Thanks, kbenson.

30% of the populace has 3+ accounts? That doesn't make sense - it doesn't follow power-law distribution.

Most like the .01% have hundreds to thousands of accounts. What does that imply?

If somebody has hundreds or thousands of accounts, it implies they are running bots ... which I guess brings the whole argument back to the point that active humans with multiple personas are not responsible for inflation of reach stats.

You're implying that identities created using persona management software [1] are bots? I guess you could though they'd likely pass a turing test.

[1] https://www.dailykos.com/stories/2011/2/16/945768/-

I'm implying that if you have hundreds of accounts, you must be using software to manage it and aren't actively engaged with all of them yourself. Therefore, they are no better than bot accounts.

Well, to correct it entirely, you should probably use thirds instead of 30%, since that was either an allowance for simplicity or an error originally. I believe the correct answer is just under 3 accounts per person in the group with multiple accounts.

Alternative accounts are much more common on sites like Twitter and Instagram. I don't think it happens as frequently on Facebook.

That would be because Facebook actively makes it difficult for you to have multiple accounts.

It's not amusing to me that fake numbers can be used to justify a higher valuation, especially when those fake accounts are used to scam or mislead the "real" ones.

At some point should there be regulation on what terms like MAU means?

I don't think MAU numbers are being used to justify valuations. They may inform analysts' understanding of growth trends, in general. But, most analysts are looking at revenue, revenue/user, and margins.

This is the classic foible of calling for the mere existence of some regulation and assuming that is the same thing as the world really behaving according to some good rule.

Right now, investors researching Facebook stock have to be aware that published metrics include botspam. Suppose an authority imposed a regulation to force Facebook to publish "bot free" metrics.

Then what will happen? Legal will impose a demand on engineers, and the engineers will scratch their heads and do what it takes to satisfy legal, and legal will scratch their heads at the result and write up whatever report it takes to satisfy the regulator.

In the end, the relationship of the published number to reality will have changed in some difficult to predict, probably undocumented, way. Will investors then really have an easier time doing their research?

They haven't really been caught on anything. These aren't ad measurement results saying they did reach more people than exist, this is a reach estimation service. It's a technically complex problem that needs to take an arbitrary targeting spec with boat loads of dimensionality and return a reach estimate in 100s of milliseconds. It's based on a sampling method, of course. The reach estimation tool is a planning tool for approximations and nothing more. That's why advertisers aren't making a big fuss.

It's not that challenging to present real stats on active accounts.

OTOH, it's pretty tempting to use numbers already pre padded for you.

Some bots are not technically bots.

There may be real people behind them, but they are "employed" by various agencies as proxies for RT/likes. Some do it manually, others employ automation.

Also, many of the more prominent accounts are almost fully scripted - these people aren't really "there". Here's an example of a Salesforce exec (Verified account, btw) employing "automation":

https://twitter.com/iTrendTV/status/738375663229505536

We've done some twitter fraud analysis in the past, and it's terrifying.

Exactly. I know some people that are indistinguishable from bots even in a face to face conversation.

Ha. But the bots aren't viewing advertisements, which is what matters to shareholders.

... here's a current example. Just saw this promoted tweet from Prudential on my timeline. It's already generating engagements (10 RTs, 38 Likes):

https://twitter.com/Prudential/status/918475795479351296

Click on the Likes below the tweet (will show you list of accounts engaged), and tell me this is real human engagement.

You will see accounts like this:

https://twitter.com/Electriflyy

Twitter won't consider this a bot, yet this "person" already liked 192,000 tweets:

https://twitter.com/Electriflyy

Here's another account who engaged with this promoted tweet, for which Prudential is getting billed:

https://twitter.com/JonTheGr8

Again, look at the number of tweets this "person" liked (72,000).

Other "engagements" come from:

https://twitter.com/RobertT70111579 https://twitter.com/TeghanJustin https://twitter.com/brandon_gorsuch ... Notice the pattern?

Prudential is getting ripped off right in front of us.

This is happening now across all brands, at massive scale.

I like/reply most ads I get so they waste money on nothing. That's my little fuck you to the ad industry.

Looking at that RobertT account I wouldn’t be surprised if it’s one of the Russian IRA accounts.

Facebook is defrauding advertisers by including ad block users in their metrics!

Bots generate engagements.

"Engagements are new followers from Promoted Accounts and/or clicks, retweets or likes on Promoted Tweets. You will never be charged more than your maximum bid and you usually pay less."

https://business.twitter.com/en/help/overview/twitter-ads-gl...

The bulk of all engagements now are fake/automated.

P.S. as a bot, you don't need to "see" a tweet to engage.

How does accounting reconcile users who have ad blocking enabled? Do they even bother?

Now if you can trick the bots into viewing the ads and change their behaviour...

> How about the SEC opens an investigation into whether Twitter is defrauding investors/advertisers

The SEC protects investors, not advertisers. When non-securities fraud gets companies in trouble with the SEC, it's because said companies failed to disclose the fraud or the risk of the fraud to their investors.

Facebook discloses their "advertising revenue could also be adversely affected by a number of other factors, including...the availability, accuracy, and utility of analytics and measurement solutions offered by us or third parties that demonstrate the value of our ads to marketers, or our ability to further improve such tools; adverse legal developments relating to advertising, including legislative and regulatory developments and developments in litigation; decisions by marketers to reduce their advertising as a result of adverse media reports or other negative publicity involving us, our advertising metrics, content on our products, developers with mobile and web applications that are integrated with our products, or other companies in our industry..." [1]

[1] https://www.sec.gov/Archives/edgar/data/1326801/000132680117...

Disclaimer: I am not a lawyer. This is not legal advice. Don't commit fraud.

I think the bigger challenge would be convincing the SEC that including the bots in their user metrics constitutes fraud.

The bots count towards ad revenue, so add prior knowledge by Twitter, and boom... fraud.

Yeah, but they are still real accounts that exist. AFAIK, whether they're operated by humans or machines isn't a consideration when reporting user numbers.

But your comment almost says it: Whether or not they're actual users isn't a consideration when reporting user numbers.

Are they reporting user numbers or account #s? Is it ok if they generate 1billion accoints themselves? Why/why not? What's different?

Impressions are based on "eyeballs" on ads. Machines look at ads. Customers can project ad blocker %.

Intent to defraud would be clear if they generated a billion new, fake accounts.

Why is it not clear if they report metrics that suggest they have more users and higher engagement than they do?

The purpose of buying ad space is to expose your product to people as an impression or as an action. If I tell you I can sell you a service that will show your ads to 10 people and you agree to give me money for that statement, but 5 of those people are running 2 accounts each, that's misleading and, dare I say, fraudulent.

You're wading into "plausible deniability" territory.

Bot use endpoints that don't return ads so I don't see how they could cause fraud.

The current US administration and Twitter/Facebook bots are in your standard strange bedfellow relationship. The GOP benefits from divisive [0][1] speech not unlike that employed by bot accounts. As long as rhetorically the bots share the same intentions as the administration, namely: divisiveness and promotion of nationalistic views, there will be minimal effort in chasing down the full extent to this defrauding.

The intentions are clear - what will happen is not. The problem is you cannot prove malicious inaction.

[0] https://newrepublic.com/article/125952/gop-party-fear

[1] http://thehill.com/homenews/senate/352768-gop-senator-russia...

Both parties benefit from divisive speech. It allows them to argue about purity of ideas instead of the ideas themselves which is _fantastic_ for party loyalty.

It's not a GOP- or Democrat-specific problem.

Plus, it's not like there's only "right-leaning" bots on Twitter. Back during the election there were a couple times I tweeted something negative about DJT and my tweet was instantly retweeted and liked thousands of times by blatantly "left-leaning" bots. They hit up both sides of the aisle to stir up drama. It's practically their stated goal.

There are far, far, far more 'right-leaning' bots on Twitter than anything else. As someone who was replied to by a 'real' account connected to the bots, I got to watch the waves of fake retweets and likes on a regular weekly schedule, often a hundred or more on a Saturday morning around 2 or 3am. And this on tweets that were 2 to 3 weeks old already. It's an amplification technique to make it seem like more people support the views. You can watch it happen live in the replies to many of Trump's own tweets with accounts with names that are a combination of the words, trump-maga-america-patriot-red-right and mom or veteran or whatever and often a random number. Often accompanied by a profile picture lifted from elsewhere online. These accounts have 30k+ followers and generally post nothing except pro-Trump memes. Some do nothing except like and retweet from other areas of the botnet.

More than the US political parties it is foreign adversarial nations like Russia that benefit from dividing the US electorate.

When two guys fight they both get a black eye. It is the third guy who got them to fight that gains the advantage over them.

> Both parties benefit from divisive speech.

This is correct until you ask "how much" does one benefit from divisive speech.

There is one nationalist branch of the Owner's Party of America, and the other is Democrat.

You mean to say all those anti Trump bots replying to his tweets are doing him a favour?

I mean to say all of those pro Trump bots organizing rallies for him in America out of Russia ;) [0]

[0] http://thehill.com/policy/technology/351557-russian-actors-m...

Ever noticed how Russian groups creating Facebook pages for fake pro-Trump rallies that don't exist and don't have any attendees is "Russian groups organized pro-Trump rallies on Facebook", but them creating fake (say) BLM events that don't exist and don't have any attendees is "Russia fakes black activism on Facebook to sow division"?

Also, it requires a rather curious definition of "bots" to include actual people manually doing stuff like this. One that's political rather than technical.

Didn't the same problem show up at Reddit where they were capping the visible user numbers of certain subreddits? They were showing massively higher numbers on their advertisers page than the subreddit front page.

So either they're inflating numbers and lying to advertisers... or they're capping numbers and lying to make the "deplorable" subs look less popular.

Not too hard to prove, just need to know where to look.

The problem isn't just bots, but fraud in general.

Reply on news.ycombinator.com