Dynamic Creative Optimization in Online Display Advertising

Abstract

The large growth of online advertising and the associated growth in personalized data allows advertisers to customize their advertisements for specific customers. In particular, advertisers can customize the product that is displayed in the ad as well as the creative look of the ad. We focus on the dynamic creative optimization (DCO) problem, wherein the objective is to determine which product to show and creative to use, while constrained by business rules on advertising fatigue, advertisement retargeting, and user diversity. We study the DCO problem from an offline and an online perspective, as the number of user arrivals can be estimated but is uncertain in practice. We show that the offline DCO problem can be solved optimally using a network flow reformulation, and we develop an asymptotically optimal algorithm for the online DCO problem. Finally, in computational experiments, we show that our algorithm performs strongly on data from a simulation calibrated by practice.

Keywords: Online Advertising, Online Matching, Approximation Algorithms