While the technology to sequence multiple ads in a single ad space during a user’s visit to a webpage is available, the practice followed – for the most part – is to show the same ad in the ad space for the entire visit. In this paper, we address a challenge faced by online ad networks in managing multiple fading ads (or fads) during a user’s visit to a webpage. A fad is an ad that disappears if the user does not click on it for some length of time. The withdrawn ad could be replaced by another ad. The goal of the ad network is to determine the sequence of ads to be shown in a single ad space to the user to maximize the expected revenue generated – under a cost-per-click model – over the user’s visit. Thus, in addition to the question of the set of ads to show, the problem also seeks the sequence in which the ads are displayed and the durations for which they are shown. We consider two factors that affect the probability of a click on an ad during a user’s visit:
(i) Sojourn Effect: Influence of the passage of time.
(ii) Exposure Effect: Influence of the number of prior exposures of the ad to the user during that visit.
We provide simple and optimal policies for the ad sequencing problem when either of these two effects dominates. For the general case in which both the effects are significant, we offer a simple heuristic policy with an attractive performance guarantee. The following three enhancements to the basic sequencing problem are also analyzed:
(a) The dynamic availability of new ads – as the user’s visit progresses – that can also be considered for display.
(b) The consideration of both click ads (that generate revenue for the ad network only through clicks) and display ads (that generate revenue only through exposures).
(c) The presence of a publisher-imposed restriction that the expected revenue in each time slot exceed a certain threshold.
Session 5B, paper 2