Monotonicity in Bayesian networks
UAI '04 Proceedings of the 20th conference on Uncertainty in artificial intelligence
Model Averaging for Prediction with Discrete Bayesian Networks
The Journal of Machine Learning Research
AdWords and Generalized On-line Matching
FOCS '05 Proceedings of the 46th Annual IEEE Symposium on Foundations of Computer Science
Predicting clicks: estimating the click-through rate for new ads
Proceedings of the 16th international conference on World Wide Web
Budget optimization in search-based advertising auctions
Proceedings of the 8th ACM conference on Electronic commerce
Optimal delivery of sponsored search advertisements subject to budget constraints
Proceedings of the 8th ACM conference on Electronic commerce
"Ideal Parent" Structure Learning for Continuous Variable Bayesian Networks
The Journal of Machine Learning Research
Hi-index | 0.00 |
To participate in sponsored search online advertising, an advertiser bids on a set of keywords relevant to his/her product or service. When one of these keywords matches a user search string, the ad is then considered for display among sponsored search results. Advertisers compete for positions in which their ads appear, as higher slots typically result in more user clicks. All existing position allocating mechanisms charge more per click for a higher slot. Therefore, an advertiser must decide whether to bid high and receive more, but more expensive, clicks. In this work, we propose a novel methodology for building forecasting landscapes relating an individual advertiser bid to the expected click-through rate and/or the expected daily click volume. Displaying such landscapes is currently offered as a service to advertisers by all major search engine providers. Such landscapes are expected to be instrumental in helping the advertisers devise their bidding strategies. We propose a triply monotone regression methodology. We start by applying the current state-of-the-art monotone regression solution. We then propose to condition on the ad position and to estimate the bid-position and position-click effects separately. While the latter translates into a standard monotone regression problem, we devise a novel solution to the former based on approximate maximum likelihood. We show that our proposal significantly outperforms the standard monotone regression solution, while the latter similarly improves upon routinely used ad-hoc methods. Last, we discuss other e-commerce applications of the proposed intermediate variable regression methodology.