Bidder Preference Modeling in Sponsored Search and Applications to the Dutch Flower Auctions

Sponsored search, the mechanism responsible for the ads on the side of an internet search results page, is a primary source of revenue for search engines. Behind the scenes, advertisers place bids to have their ads shown on many different keywords, and the search engine determines which ads are displayed in which position on the page. The search engine does not know advertisers' true valuations for receiving clicks, but it could benefit from learning this information, as it could inform auction design decisions or allow the search engine to recommend new keywords to advertisers.

In this talk I will present our work on estimating advertiser preferences in sponsored search. The approach consists of two components: first we infer advertiser preferences from their observed bids, given assumptions about the rationality and beliefs of the advertisers; we then infer advertiser preferences based on the features of the auction, which doesn't require the advertiser to have been bidding on the keyword. We will conclude with a discussion on how this research may be applied to bidder preference modeling and mechanism design for the Dutch flower auctions.