Ethics: The Role Of Fraud In Internet Advertising

CHAPTER-1

Introduction

Click Fraud is a type of damage or fraud that occurs on the internet Pay per click when someone, or automated Script, or computerized program that imitates as an authorized user of a web browser clicking on an ad.

1.1 Outline

Pay per click: It is an internet advertising technique where each click of the user on the advertisement is cashed by the publisher (owner of the website). It’s an arrangement in which webmasters i.e. operators of websites, acting as publishers, which displays clickable links in exchange for charge. For every click of the user an advertisement is linked. As the Internet Advertising has evolved, a number of advertising networks have developed which act as middlemen between the publisher and advertiser.

It also includes an automated machine learning module which runs every hour and creates a probability of fraud which is then provided to the human Fraud Ops team. If the customer is new and the probability is high enough, the account will be paused for a specified number of hours to allow the human Fraud team time to review the customer account. The Fraud workbench is designed in a similar fashion to the very large-scale click filtration system in that it is rules based and each decision is made visible to the Fraud Ops team. The Fraud Ops team can in turn decide to initiate action against a suspected fraudster. adCenter deploys crawlers to publisher sites to analyze their content and operation and compare against data collected in the course of serving ads. These crawlers analyze everything from site keywords, to looking for deceptive practices and links to other known fraudsters.

Packet sniffers (internally known as “MH logs”) are also described in the adCenter Description of Method are special weblogs that sample 100% of the HTTP protocol headers in the request received by adCenter. This allows for extremely deep analysis into the origin of the traffic as well as the likelihood of it being produced by automated processes. Packet sniffers operate automatically to collect detailed information on a sample of traffic, and can also be configured to collect all records from particular

Conversion rate gives a sense of how often visitors actually react on a given ad.

4.1 Algorithm

Viceroi Algorithm uses the following steps:

For each publisher-user pair, viceroi computes the log of the sum of ad click revenues generated by the given user on the publisher’s site.

For each publisher, viceroi sorts the per user log revenue sums and retains a vector of N quantile values. Recalls that quantile values are sampled at regular intervals from the probability distribution function (PDF) of a random variable.

For the base line, viceroi computes the point-view average of the quantile vectors for the given set of ethical publishers.

Finally for each publisher viceroi computes the point wise difference between the publisher quantile vector and the baseline quantile vector. The publisher click spam score is simply the L1 norm of the difference vector. In the online component, whenever an ad is clicked viceroi checks if the publisher is flagged and the user clicking the ad falls in the flagged quantile region. If so, the click is