Robust ad delivery plan for guaranteed display advertising

In this paper, we study the ad delivery problem of guaranteed display advertising. From perspective of publishers, we allocate inventory of page views, which is uncertain, from multiple user segments to multiple ad campaigns. Specifically, each ad campaign targets some user segments and has a certain demand of page views, which is guaranteed to be fulfilled by the publisher. We propose a desirable ad delivery plan with high audience reach and robustness. High audience reach brings high brand awareness for advertisers, meanwhile, high robustness makes the plan being immunized to the uncertainty of ad inventory. We derive a closed-form measure of expected audience reach and then build a chance-constrained programming model. To resolve the computational difficulty encountered in instances of large scale, we aggregate the audience space by clustering user segments. Numerical experiments show that near-optimal plans can be found by our approach, and its performances in audience reach robustness are highly acceptable.