Finite-time Analysis of the Multiarmed Bandit Problem
Machine Learning
Externalities in online advertising
Proceedings of the 17th international conference on World Wide Web
Exploration-exploitation tradeoff using variance estimates in multi-armed bandits
Theoretical Computer Science
Hybrid keyword search auctions
Proceedings of the 18th international conference on World wide web
Characterizing truthful multi-armed bandit mechanisms: extended abstract
Proceedings of the 10th ACM conference on Electronic commerce
The price of truthfulness for pay-per-click auctions
Proceedings of the 10th ACM conference on Electronic commerce
Explore/Exploit Schemes for Web Content Optimization
ICDM '09 Proceedings of the 2009 Ninth IEEE International Conference on Data Mining
Maintaining equilibria during exploration in sponsored search auctions
WINE'07 Proceedings of the 3rd international conference on Internet and network economics
Sponsored search auctions: an overview of research with emphasis on game theoretic aspects
Electronic Commerce Research
Ad click prediction: a view from the trenches
Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining
Machine learning in an auction environment
Proceedings of the 23rd international conference on World wide web
Quizz: targeted crowdsourcing with a billion (potential) users
Proceedings of the 23rd international conference on World wide web
| Hi-index | 0.00 |
The standard business model in the sponsored search marketplace is to sell click-throughs to the advertisers. This involves running an auction that allocates advertisement opportunities based on the value the advertiser is willing to pay per click, times the click-through rate of the advertiser. The click-through rate of an advertiser is the probability that if their ad is shown, it would be clicked on by the user. This quantity is unknown in advance, and is learned using historical click data about the advertiser. In this paper, we first show that in an auction that does not explore enough to discover the click-through rate of the ads, an advertiser has an incentive to increase their bid by an amount that we call value of learning. This means that in sponsored search auctions, exploration is necessary not only to improve the efficiency (a subject which has been studied in the machine learning literature), but also to improve the incentive properties of the mechanism. Secondly, we show through an intuitive theoretical argument as well as extensive simulations that a mechanism that sorts ads based on their expected value per impression plus their value of learning, increases the revenue even in the short term.