Kurzbeschreibung
Sponsored search has grown to be an important share of the advertisement market and a major income source for large internet companies. Its success relies not only on the explosive internet use but also on successful implementation that made it a highly efficient and profitable concept for advertisers and search engines, in particular, through the use of keyword auctions. Naturally, it also attracted significant interest from the research community, where the focus is on modeling and analyzing the sponsored search system using the tools and methods of game theory . Up until now, most of the research relies on the simplest models to extract useful results. Unfortunately, practical scenarios present a variety of complicated parameters that lead us to question the practical significance of these models. We extend the most common models to encapsulate an aspect of the system that only recently has begun to draw attention; the effect of competing advertisements on the user's actions and subsequently on the advertisers' campaign efficiency. As most of these online campaigns are run in a highly competitive environment, for example when the user has to make a unique purchasing choice, it is known that nearby advertisements impose complicated correlated effects on each other's performance. We present models that take these effects into account while remaining simple enough for us to answer the most basic game-theoretic questions about them such as the presence of equilibria and their efficiency. We also compare our models to the most common model and show their significant advantages. We follow our modeling analysis by investigating another approach for improving the current implementations. Although search engines are experimenting with different forms of advertisers' bidding, all of them only allow advertisers to bid on a single auction outcome such as impressions, clicks or conversions. We propose a mechanism that allows the advertisers' to bid in all interested outcomes in a combined fashion. We show that this mechanism has analytical advantages and also showcase through simulations that it performs at least as well as the typical mechanism in the most interesting metrics.