How does Google's antifrood system: Tuzhilin report

In connection with a number of lawsuits, recently the topic of click fraud has become particularly relevant. Several analytical companies published the results of their own research. They estimate the level of fraud on the largest search engines at 14, 1-14 , 6%.

During the Lane's Gifts against Google trial, the court asked independent experts to carefully check Google for tricks with fake clicks. The assessment was conducted by a professor at the New York University Information Systems, Dr. Alexander Tuzhilin , who, while working on the document, visited the company's office three times and talked with more than a dozen Click Quality brigade staff (founded in 2003). A 47-page report prepared by him, Google posted online (PDF file).

In his report, Tuzhilin assesses the effectiveness of Google’s anti-fraud filters as “reasonable” (reasonable). He also notes that the quality of Google’s anti-fraud systems is constantly improving. The qualification of Click Quality employees is beyond doubt. Google’s current anti-fraud filter suite is fairly stable and only requires periodic configuration. The system has four levels of protection with a different algorithm of work in each of them: pre-filtering (clicks of Google employees and erroneous entries in the logs), online filtering (differences from normal user behavior, inconsistency with the specified rules for a normal click), automatic system of offline recognition and manual offline fraud recognition system. There is no need to radically redo this system, Tuzhilin believes, even if massive fraud attacks begin on the AdWords network, as it happened several times during the past year, it will cope. Moreover, it is practically impossible to drastically improve it: each new filter in the online filtering system is able to recognize a maximum of 2-3 % of false clicks that are not recognized by other filters.

From the report of Tuzhilin, you can find out some interesting details about how Google’s anti-fraud system works. For example, its important element is the collection and analysis of information about the actions of the user on the advertiser's site after he went there through an advertising link. In particular, the advertiser’s website marks a “key” page (for example, a shopping cart in a store), a visit to which is also recorded.
')
There is a report and criticism. According to Tuzhilin, the work of Google’s anti-fraud system is still not ideal. For example, as a weak spot, he notes the fact that the system records the user's behavior only after the moment he followed the advertising link. But his activity up to this point is not taken into account.

In addition, Google does not show advertisers which transitions to the links it considered normal and which ones to be wrong. In the company's reports there are statistics by day. More recently, the AdWords system indicates the number of fake clicks and their percentage of the total number, but no more.

Alexander Tuzhilin sincere bewilderment was caused by the fact that Google only in March 2005 recognized double clicks on advertising links, which were previously counted as two transitions, wrong. This seemingly obvious case of a fake click was considered normal in Google for several years, and after adding the appropriate filter, the company, by its own admission, lost a “substantial part” of advertising revenue. Such cases make you wonder: Is Google really sincerely struggling with Freyud in its advertising network? Actually, in his report, Tuzhilin listed a number of anti-fraud methods (p. 34), which for some reason does not use Google.