| Internet advertising is crucial for the thriving of the entire Internet, since it allows producers to advertise their products, and hence contributes to the well being of e-commerce. Some publishers are dishonest, and use automation to generate traffic to defraud the advertisers. Similarly, some advertisers automate clicks on the advertisements of their competitors to deplete their competitors' advertising budgets. Moreover, some dishonest publishers and advertisers form coalitions to circumvent the fraud detecting techniques.In this thesis, we describe the advertising network model, and discuss the issue of click fraud that is an integral problem in such setting. We explain the classical approach of detecting click fraud, which judges publishers' and advertisers' traffic based on how the advertisements behave on the publishers' sites. We describe the drawbacks of the classical approach, and propose using streaming and sampling algorithms on aggregate traffic as a viable way of detecting automated traffic, and explain how the traffic analysis approach complements the classical approach. We start by classifying the click fraud techniques into two major classes based on the motivation of the fraudulent publishers and advertisers. We devise traffic analysis algorithms that detect both classes of fraud attacks. This motivated us to solve several data analysis problems, such as detecting frequent stream elements, finding association rules between elements in a stream, and sampling large sets to find similar pairs of sets among numerous sets. We conclude with some successful findings of our attempt to detect fraud on a real network. |
