Fraud Prevention in Online Digital Advertising

In this chapter, we briefly introduce the computational advertising, including search advertising and display advertising. We explain the reality of fraud in digital advertising, and summarize types of fraud methods commonly observed in the industry for making illicit returns.

In this chapter, we briefly describe the digital advertising ecosystem, mainly from the display advertising perspective. We will first describe the real-time bidding framework for online digital advertising, including technical platforms for publishers, advertisers, and the market place for online Ad inventory buying and selling. After that, we will describe major business model of online advertising, and explain three types of revenue models commonly used in Ad systems, including impression-based revenue model (CPM), click-based revenue model (CPC), and action based revenue model (CPA).

In this chapter, we first propose a taxonomy to summarize fraud in online digital advertising. The taxonomy provides a complete view of major fraudulent activities in answering questions related to Who does What to Whom, and How. The proposed fraud taxonomy includes three major categories: placement oriented fraud, network traffic oriented fraud, and action oriented fraud. Placement oriented fraud mainly intends to manipulate or modify publisher pages or modify the web pages showing on the user’s devices in order to increase the number of impressions or clicks. Traffic oriented fraud generates fake traffic to inflate the number of impressions or clicks generated from individual sites or placements. Action oriented fraud aims to target users’ actions in order to generate revenue.

This chapter provides a comprehensive review of Ad fraud in three major categories: placement fraud, traffic fraud, and action fraud, which are at different levels of online advertising. Placement fraud mainly focuses on the pages which displaying the Ads. For placement oriented fraudulent activities, they often modify publisher pages or the web pages showing on the users’ devices to increase impressions or clicks. Traffic fraud mainly tries to manipulate the network traffic to inflate the number of impressions generated from individual sites or placements. Action fraud targets users’ meaningful business actions, such as filling an online form or survey, completing an online purchase order, or use users’ previous actions or behaviors to re-target valuable customers. For each type of fraud, we will also review detection methods and approaches for online Ad fraud prevention.

In this chapter, we discuss measures and benchmark datasets commonly used for Ad fraud detection. The measures include fraud detection accuracy, precision, recall, F-measure, and AUC scores which are commonly used to validate the performance of classifiers for classification. In addition, we also summarize several real-world datasets which are currently available for Ad detection and computational advertising research in general.

This chapter reviews both commercial Ad fraud detection and prevention systems and the ones developed in academia. For commercial systems, they mainly emphasize on the efficiency, so fraud detection can be achieved at pre-auction level (e.g. less than 10 ms). The systems developed in academia are often more sophisticated in their designs and mathematical models. Yet the efficiency of such systems for online usages are often not strictly evaluated.

Online advertising fraud represents a significant portion of deceiving actions in digital advertising systems which use numerous technologies to derive illicit returns. Even the most conservative estimation has shown that more than 10% of Ad inventory is consumed by bot or fraud impressions. Despite of the fast growth of the computational advertising in modern communication networks, no comprehensive literature review or research documentation exists to summarize forms of fraud in Ad systems. In this book, we provided a comprehensive review of fraud activities in Ad systems, by using a tiered taxonomy to summarizes Ad fraud at different levels and from different perspectives. Our taxonomy categorizes Ad fraud into three major categories, including (1) placement fraud, (2) traffic fraud, and (3) action fraud, with each category focusing on publisher web sites/pages, network traffic, and user actions, respectively. Our literature review provides direct answers to key questions such as the major types of frauds in Ad systems, key approaches and characteristics of different types of fraud, major methods used to detect Ad frauds, and ground truth, measures, tools available to assess fraud and support research in this domain. This book delivers a first hand research guidance for online Ad fraud prevention. It also serves as technical reference for industry practitioners or developers to design their own fraud defending systems.