The goal of this project is to maximize the utility of Yahoo’s ad auction system by providing long-term value to all three groups of participants in search advertising - users, advertisers and publishers.
Search engine users expect to find, with limited effort, relevant organic results and, in case of commercial intent, attractive offers. Bad ads can potentially tarnish their experience with the search engine, and as a consequence keep them from clicking again on any ad, or from coming back at all. Advertisers try to maximize their return on investment, which depends on their ability to convert clicks into sales or other events that create value, the price per click, the conversion rate, and the number and price of clicks. Publishers are web sites on which the search engine results are displayed. This includes the portals operated by search engines themselves such as Yahoo and Google, and other partners that typically enter in a revenue-sharing agreement to show the results on their site.
With this project, we are seeking to ensure publisher's long-term revenue by improving user and advertiser satisfaction. We are designing modifications of the GSP auction scheme that take into account an estimate of total utility (to users, advertisers, and publishers). We are also developing online utility and relevance models, and developing models for more accurate user personalization.
These models can be applied to a variety of algorithms, including filtering of ads with presumably bad user experience, and changing ad ranking by taking into account the hidden cost of bad ads driving away users in the future. Other applications include adjusting the number of ads shown on top of web results, and more precise personalization of the search experience to the user profile.One example of an approach under investigation is the Hidden Cost Model, in which irrelevant ads are assumed to have a long term cost to the search engine due to a poor user experience, and an estimate of this cost is used in determining the rank and price of the ads. Another approach is examining the different impact on the user depending on the position of the ad on the page, and adjusting the total utility based on ad position. And a third approach is incorporating estimates of the relevance of both organic results and ads in order to influence the ad placement algorithms to deliver an optimal user experience. References Zoe Abrams and Michael Schwarz, Ad Auction Design and User Experience, Vol.2, Special Issue I: Theoretical, Empirical and Experimental Research on Auctions, Applied Economics Research Bulletin, Fall 2008 Yagil Engel and David Maxwell Chickering, Incorporating user utility into sponsored-search auctions, pp 1565-1568, 7th International Joint Conference on Autonomous Agents and Multiagent Systems, 2008