||Estimating Performance for Sponsored Search Advertisement
||Winther, Ole (Intelligent Signal Processing, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
||Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
||Sponsored search advertising has grown rapidly over the last years to become the most dominant online advertising model generating the most revenue. Monitoring and evaluating the performance of sponsored search advertisements has become more and more important for the advertisers, allowing them to assess the profitability of their marketing efforts. Documentation of the effect and profitability of online marketing activities is in ever increasing demand from management.
This thesis explores and presents methods on how to accurately estimate performance metrics for sponsored search advertisement. Important performance metrics used by the advertisers for evaluate advertising effectiveness and profitability are the click-through rate and the conversion rate. Advertisers usually use historical data to estimate these metrics. The challenge is how to provide accurate estimates for ads with little or no historical data, which includes newly created advertisements.
We show how to estimate performance for sponsored search advertisement using logistic regression combined with the features of the ads to estimate both click-through rate and conversion rate. Different methods and models for logistic regression are explored and results are compared both with regard to predictive performance and their properties. The methods that we explore are logistic regression, regularized logistic regression, sparse Bayesian logistic regression and Bayesian logistic regression with sampling. The models are trained and tested using real-world data sets from online advertising and compared with a baseline model. Our findings show that we can accurately estimate the performance of sponsored search advertisements given features of ads such as the bidded keyword, title and description.
Finally we provide an implementation of a concrete application for sponsored search advertising, allowing advertisers to create and make suggestions for new keywords and evaluate the performance using the models to compute an estimated profit metric.
||Thesis not public available.
||Technical University of Denmark (DTU) : Kgs. Lyngby, Denmark
Creation date: 2009-09-02
Update date: 2009-11-04