URL: http://www.datashaping.com/security.shtml
ABSTRACT
The problem of click fraud, invalid or poor clicks is nearly as old as
paid-per-click advertising. Only recently has it been acknowledged by major
search engine companies. It represents the most serious threat to the online
advertising industry. Although there is no formal definition to click fraud,
a good description can be found at datashaping.com/clickfraud.shtml.
We review various state-of-the-art solutions to address click validity in
pay-per-click programs, including proprietary IP blacklists, entrapment,
ad-hoc design of experiment and detection of false positives. Our core
contribution is a scoring system tied to the advertiser's ROI and matching
conversion rate distributions.
New original fraud cases will be discussed, including impression and ad
relevancy fraud, automated clicks from shareholders and political activists,
and accidental click fraud generated by email spammers.
TOPICS COVERED
Identifying unique users in server logs
Analytics: decision trees
Analytics: comparison of statistical methodologies
Analytics: tree farms and other boosted trees
Click scoring: logical scores tied to ROI and conversion rates
Metric selection: velocity, delta, binned metrics; optimum binning
How to design an alarm system
Testing ad campaigns, click fraud entrapment
Keyword purchase optimization to minimize fraud and increase ROI
Case studies
Statistical analysis of Google and Yahoo reports to detect abnormalities
How to identify fraud hiding behind AOL proxies?
Types of fraud and ad-hoc detection strategies
Advertiser training: tagging systems to correctly monitor click activity
Advertiser training: missing clicks in server logs - explanation, quick fix.
--
Vincent Granville, Ph.D.
C.E.O and President
Data Shaping Solutions, LLC
[log in to unmask]
925-759-7308 (cell)
www.datashaping.com
|