Google Says ...

An unofficial, unaffiliated source of comment and opinion on statements from Google, Google employees, and Google representatives. In no way is this site owned by, operated by, or representative of Google, Google's point of view, policies, or statements.

My Photo
Location: California, United States

Use your imagination. It's more entertaining.

Monday, July 24, 2006

Google's click-fraud prevention techniques are 'reasonable'

So says Alexander Tuzhilin, the independent 'expert' who reviews Google's fraud-detection systems for a court. It's not clear to me, after reading this document (which includes the author's bio), what qualifies him to review Google's click-fraud detection methodologies. He does not claim to have any prior experience or exposure to the management or generation of invalid clicks (as Google describes them).

Nonetheless, his conclusion is that Google makes a reasonable attempt to detect and neutralize invalid clicks. Googlers are understandably happy to be so vindicated in an official document. However, the fact that the author quotes Wikipedia further underscores his essential naivete. He provides no reservation about the votility of Wiki entries (although any comparison of a legal analysis from 100 years ago would be expected to make reference to contemporary dictionaries and source materials, he provided no publication date for the Wiki reference).

I think Google makes a better presentation in its filed objections to a proposed $90 million settlement, especially where they point out that the proposed settlement implies fraud exceeding their total revenues to date has occurred. If there are any valid scientific claims being made in this case, neither side has done a very good job of providing them.

Dr. Tuzhilin's analysis even includes a self-admitted unscientific Zipf graph, where he analyzes the long tail of invalid clicks. While there is a certain logic to what he proposes, he offers no evidence to support his contention that the behavior he is analyzing conforms to his proposed model. That would be equivalent to Einstein saying, "Well, I think Time changes near Mass, but I have no mathematical model to show how this works." It has taken scientists decades to accumulate confirmable observations that support the extensive math accompanying Einstein's theory, so maybe I'm being a little harsh in comparing the long tail model to Relativity.

However, Dr. Tuzhilin only examined the issue from Google's perspective. His report indicates nothing about any attempts to contact people who specialize in, or rely-upon invalid click generation. Such people exist, they have been in business since before Google was founded, and based on my own conversations with some members of that shadow industry, they were already using (8 years ago) advanced methodologies which Google's vague measures appear to be incapable of detecting.

Dr. Tuzhilin suggests that Google has not implemented any data mining techniques in its filter technology. Google, if I were you, I'd assign about a dozen people to catch up on that major deficiency right now. You really have no idea of what you are up against. You need to look at the history of those IP addresses that are clicking on your ads and search results. In my opinion, based on what I have read across the Web, I think Google probably does a very good job of attempting to detect click fraud. Probably no one is better at it by now.

I'll agree that their methods are reasonable, based on the available data.

But are they effective on a large enough scale to ensure advertisers that a majority of the invalid clicks are captured? There are two areas where that question must be answered: in the aggregate and in the specific. That is, overall, the effectiveness of the program may be acceptable ("acceptable" may require something more than a reasonable effort in some people's opinions). But some specific campaigns may be targeted for abuse that is slipping by the filters.

What Dr. Tuzhilin's report underscores is the fact that Google does not know how effective its filters truly are.


Post a Comment

<< Home