Something that I’ve been wondering about for a while. Google Ads don’t necessarily match their advertisements to websites in quite the manner that you’d expect, presumably because of the way that its underlying algorithm works. “Brad DeLong’s site”:http://delong.typepad.com/, for example, seems to have become the new in-spot for Republican and Conservative singles to hook up with each other, while “Nathan Newman”:http://www.nathannewman.org/log/ rather improbably provides a venue for union-busting specialists to connect with their core clientele in the business community. I wouldn’t at all be surprised if many of Nathan’s readers make a habit of clicking on the link to “Union-Free Labor Relations Training” on a regular basis. After all, each time that they do, a small sum of money presumably disappears from the advertising budget of a rather slimy organization, and reappears (after Google deducts its cut) in Nathan’s Paypal account. Now personally, I’ve no problems with that. But does this undermine the rationale behind using Google Ads for politically targetted advertising? Left-leaning blogs are likely to “sound” Republican to Google’s algorithm because of the frequency with which they mention Republican politicians (and Republican blogs will sound left-wing). Thus, they’re likely to attract a disproportionate number of ads which are aimed at exactly the wrong population. Many of the people who read these blogs are unlikely to want to click on the ads for any sincere motives. The same, of course, is true for right wing blogs harping on how horrible the Democrats are; again they’ll appear to be “good” bets to an automated algorithm for advertisements aimed at leftwingers, unless that algorithm is sophisticated indeed.
Unless Google changes its algorithm, I can’t imagine any easy technical way of distinguishing ‘fake’ clickthroughs from real ones, except in the most straightforward of cases (e.g. the same person at the same IP address repeatedly clicking on an ad again and again). Now this may in reality be a non-problem – I’ve seen no data on it- but in principle, Google Ads seems to me to be quite vulnerable to politically motivated attacks, which could prove quite expensive for the advertisers.
{ 18 comments }
Barry 06.27.05 at 2:42 pm
Henry, thanks for that advice. I’ll be a rabid clicker, from now on.
Seth Finkelstein 06.27.05 at 2:43 pm
In principle, yes. But in practice, I think few people are that motivated for political reasons. My impression is that plain old financial fraud is the big problem. And the measures to deal with that easily take in political motivations as a small subset.
ben wolfson 06.27.05 at 4:06 pm
Then there’s this greasemonkey script, which indiscriminately clicks all adwords on a given page.
foo 06.27.05 at 4:48 pm
Google knows that distinguishing “false” click-throughs from “real” ones is one of the key issues (and problems) with their online advertising business… they talk about this explicitly in some of their SEC filings and reports (in the parts where they are supposed to disclose potential risks to their business plans).
Anyway, the key mistake here is “left-leaning blogs are likely to “sound†Republican to Google’s algorithm because of the frequency with which they mention Republican politicians.” In other words, the problem your pointing out only “undermines the rationale” for word-frequency-based schemes.
There’s no reason why more sophisticated (and not , perhaps, even that much *more* sophisticated) algorithms couldn’t do a better job.
After all, your brain does it, so why not a computer?
Barry 06.27.05 at 5:03 pm
Oh, ben, you are cruel :)
y81 06.27.05 at 5:06 pm
Do you think that Republicans don’t read Brad DeLong? I don’t know of any reliable evidence as to whether most people confine themselves to reading blogs with which they are in basic political agreement. (Many commentators here vow regularly not to read Instapundit, but I don’t know whether they practice what they preach, much less what the average left/liberal blog reader does.)
I personally don’t read Brad DeLong often, because I find him puerile, but I read this site and Kevin Drum almost every day, even though I often disagree with both the political and culinary views expressed here.
yabonn 06.27.05 at 5:54 pm
Wasn’t that weirdo push-up-in-panties baldie (“master of his own weight!”) guy a pretty sure marker for right wing blogs?
justin 06.27.05 at 6:54 pm
I’ll admit that I’m no expert on Googles algorithm, but do the ads only work on keywords, or do they also take into account links? they did take into account links it is unlikely that a liberal blog like dailykos or a conservative blog like powerline would ever attract the ‘wrong’ type of advertising, as links tend rarley leave the ‘cluster’ of websites that they exist in.
in regard to a previous post, I would suspect that many readers of the ‘top’ blogs don’t go to the other side of the isle that much. I would suspect the readers of CT might as this blog is more apolitical and academic in nature…. but the average reader of any of the popular political blogs? I doubt it.
Teresa Nielsen Hayden 06.27.05 at 7:40 pm
I’m hoping for a scenario like the one that drove Aldus and Adobe to develop better and better illustration software. In this version, the weblog ad market remains stable enough for ad placement services to competitively refine their methods, gaining market share among advertisers by being able to promise them more accurately targeted ads.
The bad alternative (barring economic collapse, alien invasion, transformational new technology, or other large events that wash away the whole sandcastle) is that ad placement services blunder too visibly, advertisers take their nickels and dimes elsewhere, the bottom falls out of the weblog advertising market, and no one is left with sufficient incentive to develop more sophisticated methods.
That invisible hand has its awkward days as well as its deft ones.
Andrew 06.27.05 at 8:18 pm
Google’s algorithm is much more sophisticated than simply counting the words in the piece and displaying the adverts that have similar counts. Anyway, haven’t you ever read a short piece and been mistaken about the author’s intent? I know I have. It’s not inconcievable that a human could make such a mistake, especially if that human were not very bright. I’m not defending bad software, but as a computer scientist, I sympathise.
It’s silly to say that a computer should be able to do what a human brain can do. Sure, a computer can crunch some sorts of data much faster than a person, but with others even the best software and most expensive software fail miserably. Take as trivial a task for a human as driving, which is undertaken a billion times a day by hundreds of millions of people, with relatively few mistakes. The best robot driver in the Darpa “grand desert challenge” only managed about 7 miles on a open track in broad daylight relatively free from obstacles. Or at as seemingly straight-forward a task as playing the boardgame Go, even a novice player can beat the most sophisicated software.
A computer can certainly calculate the 100 digit of pi faster than I ever could, but I bet I will always be able to pick a better present for my mum than amazon’s suggestions.
Julian 06.28.05 at 12:53 am
My wife works on web ad placement (though not for Google). The more sophisticated algorithms, in addition to counting clicks, take into account how often those clicks materialize as a sale or other transaction.
Jeremy Leader 06.28.05 at 1:37 am
I would guess that Google does some pretty sophisticated statistical analysis to try to detect the sort of bogus clicks Henry describes. Of course, they have an incentive to let such clicks through, to pump up revenue, but they also know they have to worry about the scenario in Teresa’s 2nd paragraph, wherein they lose the advertisers’ trust. Note that large advertisers (and maybe some small ones) probably have ways of tracking which sources of clicks actually result in more or fewer sales, so they’d notice any large scale funny business.
Also, my understanding is that Google chooses ads based on expected revenue, which implies that they have some way of predicting roughly how well any particular ad will perform on any particular type of page (as classified by their algorithm). So I doubt they’d keep running ads that weren’t making them much money.
In other words, I suspect that Google (and their competitors) are busily adjusting their market mechanisms to keep that invisible hand from tipping too far one way or another.
Disclaimer: I work in advertising systems at Yahoo!; we compete with Google. However, my opinions expressed here bear no connection to my employer.
abb1 06.28.05 at 2:09 am
…but I bet I will always be able to pick a better present for my mum than amazon’s suggestions.
I don’t think it’s so obvious. I suspect that with enough information and with sophisticated enough algorithm, a computer might be able to beat you in picking present for your mom. But you do it much cheaper. For now.
SusanC 06.28.05 at 6:10 am
…but I bet I will always be able to pick a better present for my mum than amazon’s suggestions.
I agree with the general sentiment – that there are some tasks where computers aren’t going to be able to beat humans for quite a while – but I’m not sure about the example. A simple algorithm together with a very large database might be able to do much better than a human at predicting which books/music/films someone else would like. Amazon’s recommendations are often remarkably accurate (although they’re sometimes completely off the mark).
theCoach 06.28.05 at 10:05 am
If this became a real problem, wouldn’t it become viable to market lists of domains that you would not want your ads on? A liberal group could come up with their list, and a republican group theirs (a social conservative, a libertarian left wing, etc.). Comapnies placing ads could select a list provider and the list provider would get a very small cut of the premium paid for list filtering.
andrew, “always” has a bigger scope than you seem to think.
Nathan Newman 06.28.05 at 10:21 am
I’ve meant to comment on the Google Ads thing myself, but it’s actually a reasonable placement of the ad. A rapidly anti-union person or company is probably more likely to be reading my site — ie. know the enemy — than reading about garden tools, so a union buster spends their money better in pro-union sites than at a gardening magazine.
Yes, worlds are insular on the Internet at times, but many enemies read each others stuff more than third party organizations pay any attention.
But please, keep clicking away!
Jon 06.28.05 at 11:44 am
It’s not just Google’s ad algorithms that can manipulated but its humans as well.
Google’s Adwords ads are not reviewed for policy violations but on an ad hoc basis after the fact. As a result, advertisers can find their ads suddenly dropped by Google following complaints from partisans of any stripe.
For more on this disturbing situation, see:
– “Still Doing Evil:Google and Political Speech”
– “Google’s Gag Order: An Internet Giant Threatens Free Speech”
Jeremy Leader 06.30.05 at 12:17 pm
“Google Sued Over ‘Click Fraud’ in Web Ads”:
http://go.reuters.com/newsArticle.jhtml;jsessionid=CJNIHMABY5FJMCRBAEOCFFA?type=topNews&storyID=8940234
The story notes:
Click fraud is not “fraud” as defined under the law. Rather, it is an industry term used to describe the deliberate clicking on Web search ads by users with no plans to do business with the advertiser.
Also, note that the plaintiff, Click Defense, Inc., sells software to detect or prevent click fraud, which makes me think this might just be a case of publicity-by-lawsuit.
Comments on this entry are closed.