An Analysis of Autonomous Bidding Strategies in Ad Auctions


Speaker


Abstract

The Trading Agent Competition for Ad Auctions (TAC AA) is a simulated sponsored search domain that tries to capture some of the complex dynamics of bidding in sponsored search auctions.  In this paper, we present a suite of agents that we developed for the 2009 TAC AA tournament.
 
We decompose the agent's problem into a modeling sub-problem, where we estimate values such as click probability and cost per click, and an optimization sub-problem, where we determine what to bid given these estimates.  Most of our models estimate values using an ensemble of regression predictors, which we empirically show to be more accurate than any single regression predictor used in isolation.  Our optimization algorithms are composed of rules-based algorithms, which can make decisions with minimal information, and more sophisticated algorithms such as multiple choice knapsack solvers and integer linear programs, which are capable of finding better solutions than rules-based methods but require more accurate models.
 
We create two frameworks to evaluate the performance of our agents: a framework that evaluates model accuracy, and a decision-theoretic simulator, which evaluates agent performance given varying degrees of model error.  We use these testing frameworks to determine which optimization algorithm is most suitable for an advertiser with a given amount of model error, and which model improvements will lead to the greatest gain in performance.  Finally, we run controlled experiments in the actual TAC AA domain to evaluate agent performance in the game-theoretic setting.
 
This work is joint with Jordan Berg, Eric Sodomka, and Victor Naroditskiy.
 
Dr. Amy Greenwald is Associate Professor of Computer Science at Brown University in Providence, Rhode Island.  Her primary research area is the study of economic interactions among computational agents.  Her primary methodologies are game-theoretic analysis and simulation.  Her work is applicable in domains like dynamic pricing and autonomous bidding.  She was awarded a Sloan Fellowship in 2006; she was nominated for the 2002 Presidential Early Career Award for Scientists and Engineers (PECASE); and she was named one of the Computing Research Association's Digital Government Fellows in 2001.  Before joining the faculty at Brown, Dr. Greenwald was employed by IBM's T.J. Watson Research Center, where she researched Information Economies.  Her paper entitled "Shopbots and Pricebots" (joint work with Jeff Kephart) was named Best Paper at IBM Research in 2000.
 
Download slides
 
Contact information:
Dr. Wolf Ketter
Email