TY - JOUR TI - Observing and optimizing online ad assignments DO - https://doi.org/doi:10.7282/T3JS9SD9 PY - 2015 AB - The main focus of this thesis work is on optimization and observation of ad assignments in online ad markets. Online ad markets allocate billions of impressions to advertisers while satisfying an array of constraints. Their revenues support the Internet ecosystem. They highlight theory problems and inspire systems research. In this thesis work we initiate the study that seeks to understand mechanisms and dynamics of advertising markets. We develop a scalable crawling capability that allows us to harvest a corpus of ads across a large number of websites and user profiles. We establish that user profile is essential in display ad markets: 50% of observed websites have at least 80% of their ads targeted at profiles. Further, we introduce cardinal auctions for selling multiple copies of a good, in which bidders specify not only their bid or how much they are willing to pay for the good, but also a cardinality constraint on the maximum size of the allocation in which they are willing to participate. We perform the first known analyses of Price of Anarchy and revenue of cardinal auctions. Finally, we introduce a new class of online allocation problems with secondary metrics, in which the goal is to optimize one metric (e.g., revenue), while meeting another (e.g., cost of user conversion). We suggest a number of theoretical approaches to the problem and test one of them in a real-world setting by using it in ad allocation in a ad network. KW - Computer Science KW - Internet advertising LA - eng ER -