Detecting Click Fraud in Pay-Per-Click Streams of Online Advertising Networks

Abstract

With the rapid growth of the Internet, online advertisement plays a more and more important role in the advertising market. One of the current and widely used revenue models for online advertising involves charging for each click based on the popularity of keywords and the number of competing advertisers. This pay-per-click model leaves room for individuals or rival companies to generate false clicks (i.e., click fraud), which pose serious problems to the development of healthy online advertising market. To detect click fraud, an important issue is to detect duplicate clicks over decaying window models, such as jumping windows and sliding windows. Decaying window models can be very helpful in defining and determining click fraud. However, although there are available algorithms to detect duplicates, there is still a lack of practical and effective solutions to detect click fraud in pay-per-click streams over decaying window models. In this paper, we address the problem of detecting duplicate clicks in pay-per-click streams over jumping windows and sliding windows, and are the first that propose two innovative algorithms that make only one pass over click streams and require significantly less memory space and operations. GBF algorithm is built on group Bloom filters which can process click streams over jumping windows with small number of sub-windows, while TBF algorithm is based on a new data structure called timing Bloom filter that detects click fraud over sliding windows and jumping windows with large number of sub-windows. Both GBF algorithm and TBF algorithm have zero false negative. Furthermore, both theoretical analysis and experimental results show that our algorithms can achieve low false positive rate when detecting duplicate clicks in pay-per-click streams over jumping windows and sliding windows.