A New Ranking Scheme of the GSP Mechanism with Markovian Users

Sponsored search auction is used by most search engines to find ads shown the web page of search results. The income of this targeted advertising business is a big part of the revenue of most search engines. The most widely used approach to find ads is called Generalized Second Price (GSP) auction. Most previous works about GSP auction are based on the separation assumption: the probability a user will click on an ad is composed of two independent parts: a quality factor of the ad itself and a position factor of the slot of the ad. The previous model does not include the externality an ad may bring to the other ads. We study a GSP auction in a Markovian user model where the externality is considered by modeling a user’s probability behavior when he reads ad list. In particular, we propose a new ranking scheme for the bidders. We prove Nash equilibrium always exists in the auction and study the efficiency of the auction by theoretical analysis and simulation. We compare our results with social optimum and previous approaches. Comparison shows that our scheme approximates the social optimum and improves previous approaches under various circumstances.