Do you like the ads you see on Bing? Are they useful or relevant to you? As an advertiser, are you getting valuable user clicks? Do you feel you are paying a fair price per click?
We attack the challenge of computational advertising by applying our research in machine learning and mechanism design to the two following areas:
- Click prediction: In paid search (or sponsored search) the advertiser is not charged when their ad is shown, but only when a user clicks on the ad. The predicted probability that a user clicks on an ad impression (or CTR for click-through rate) is therefore a crucial quantity for optimally allocating ads to a page. Furthermore, the predicted CTR also affects the amount the advertiser is charged per click. Accurate predictions benefit both parties involved in the paid search marketplace: the user who sees more relevant ads and the advertiser who gets more clicks at a fairer price.
[Our team created the CTR prediction algorithm that launched with Bing in June 2009.]
- Marketplace analysis and design: We study the impact of the specific design of the paid search auction on the health and efficiency of the marketplace. Our goal is to ensure that the value for the advertiser value of every click is as high as possible: a healthy marketplace maximizes the total advertiser welfare. We build economic models and use them to make inference based on terabytes of historical bidding.
- Ad recommendation: What if we could recommend you an ad you might be interested in based on the ads people similar to you might be interested in, and based on your social network? What would be a fair pricing scheme for recommended ads?
- Matchbox: One of the greatest challenges in the web is to connect people with relevant content or products. In our Matchbox project we pursue research in large-scale recommendation systems. Check out an application of Matchbox (/en-us/projects/matchbox/default.aspx) in our joint FUSE/MSR content filter Project Emporia (http://www.projectemporia.com/).