Book Description

Since the turn of the century more and more of people's information consumption has moved online. The increasing amount of online content and the competition for attention has created a need for services that structure and filter the information served to consumers. Competing companies try to keep their customers by finding the most relevant and interesting information for them. Thus, companies have started using algorithms to tailor content to each user specifically, called personalization. These algorithms learn the users' preferences from a variety of data; content providers often collect demographic information, track user behavior on their website or even on third party websites, or turn to data brokers for personal data. This behavior has created a complex ecosystem in which users are unaware of what data is collected about them and how it is used to shape the content that they are served. In most cases personalization is useful for the users but there have been articles in the press with worrying examples of price discrimination or the Filter Bubble Effect. While this has triggered some awareness among the general public, it also made users realize how little control they have over their data and the form of the web they are presented. Meanwhile legal scholars and policy makers expressed concerns about algorithms' power to systematize biases and reduce accountability. Unfortunately, detecting the negative consequences or measuring large-scale effects is in practice very challenging, as we still lack the tools and techniques for it. My work starts with developing a methodology that will allow me to investigate personalization on any chosen content-based web service. With the help of this methodology I measure personalization on several services in two large sectors, search engines and e-commerce sites. In my investigation about search engines I find that, on average, 11.7% of results show differences due to personalization on Google, while 15.8% of results are personalized on Bing, but that this varies widely by search query and by result ranking. I also investigate the user features used to personalize on Google Web Search and Bing. Surprisingly, I only find measurable personalization as a result of searching with a logged-in account and the IP address of the searching user. Next, to further investigate location-based personalization, I design a new tool that is able to send queries to the Google Search API appearing to come from any given GPS coordinate. Assessing the relationship between location and personalization is crucial, since users' geolocation can be used as a proxy for other demographic traits, like race, income, educational attainment, and political affiliation. Using this methodology, I collect 30 days of search results from Google Search in response to 240 different queries. By comparing search results gathered from 59 GPS coordinates around the US at three different granularities (county, state, and national), I am able to observe that differences in search results due to personalization grow as physical distance increases. However these differences are highly dependent on what a user searches for: queries for local establishments receive 4-5 different results per page, while more general terms exhibit essentially no personalization. Finally, I turn my attention to personalization on e-commerce sites. Personalization on e-commerce sites may be used to the user's disadvantage by manipulating the products shown (price steering) or by customizing the prices of products (price discrimination). I use the accounts and cookies of over 300 real-world users to detect price steering and discrimination on 16 popular e-commerce sites. I find evidence for some form of personalization on nine of these e-commerce sites. I also create fake accounts to simulate different user features including web browser/OS choice, owning an account, and history of purchased or viewed products and identify numerous instances of price steering and discrimination on a variety of top e-commerce sites. Tied together, these results present the first steps towards quantifying the prevalence of personalization in web-based content services and understanding the algorithms behind them. My work also provides a novel methodology that can easily be adapted by researchers who want to study content-based web services, or regulators whose goal is to audit algorithms.