Google's war on invalid clicks

Paper includes a section on publisher terminations

[europeforvisitors]

On the AdWords forum, Whoisgregg has posted a link to a 47-page paper on how Google detects and deals with invalid clicks. The paper includes a section on publisher terminations and reinstatements.

The thread is at:

[webmasterworld.com...]

The paper itself, by Dr. Alexander Tuzhilin of NYU, is a PDF file at:

[googleblog.blogspot.com...]

And in case you aren't familiar with Whoisgregg, he's a senior member whose profile is at:

[webmasterworld.com...]

[jimbeetle]

In my reading of the paper (sections 9.2 through 9.4), I get the impression that clicks are considered to be valid until found to be invalid. If this is the case, then we should never see a "click dump" of newly classified valid clicks. What we should see upon occasion is for those publishers who have received warnings to have the number of clicks reduced. I haven't happened to see or hear of any of these instances.

[ken_b]

Clicks being considered valid until proven to be invalid doesn't mean that suspicious clicks would be published to our stats before the investigation was complete.

Not publishing suspect clicks before they were fully investigated would seem to be much more logical.

That would almost inevitably lead to a click dump sooner or later.

I don't recall anything in the paper that indicated otherwise.

[BigDave]

I never thought of click dumps as being part of the invalid click checking. It behaves totally like a server syncronization issue.

Think about it, have you ever noticed a click get credited to a channel before it gets credited to the total?

Have you ever noticed getting an "impression dump"? What would be the fraud-detection value in holding back the number of impressions?

The biggest click dumps are within 24 hours after google takes the adsense report sites down for maintenance.

The reason that he didn't mention it in the report is that when the clicks are reported to the publishers has nothing to do with when they are counted and filtered. So there is no point in mentioning it in a case involving advertisers, not publishers.

[Tapolyai]

What fascinates me about this is how Google never had so say if Mr. Tuzhilin's research was, or was not correct, nor did Google have to disclose the actual process.

[BigDave]

What fascinates me about this is how Google never had so say if Mr. Tuzhilin's research was, or was not correct,

Since this is an expert report for the plaintiff, the closest you could get is google filing an objection to it. If the report is more favorable than it could have been, and they have already agreed to settle, why would they object?

or did Google have to disclose the actual process.

If they or the report had disclosed anything that they considered trade secret, the report would have been filed under seal and we never would have gotten these tidbits.

Reading the report, it seems like they did reveal some trade secrets under NDA for evaluation, such as what the various filters do, but there was no need to include the dettails in the report.

[gregbo]

Reading the report, it seems like they did reveal some trade secrets under NDA for evaluation, such as what the various filters do, but there was no need to include the dettails in the report.

I thought the fact that they don't use advanced techniques such as machine learning, while not exactly a trade secret, was something that G might not have wished to make public.