Web publishers must balance two objectives: how to keep users engaged by directing them to relevant content, and how to properly monetize this user traffic. The standard approach is to solve each problem in isolation, for example, by displaying content that is tailored to the user’s interests so as to maximize clickthrough rates (CTR), and also by building a standalone ad serving system that displays ads depending on the user’s characteristics, the article being viewed by the user, and advertiser-specified constraints. However, showing the user only those articles with highest expected CTR precludes the display of some ads; if the publisher had previously guaranteed the delivery of a certain volume of impressions to such ads, then underdelivery penalties might have to be paid. We propose a joint optimization of article selection and ad serving that minimizes underdelivery by shaping some of the incoming traffic to pages where underperforming ads can be displayed, while incurring only minor drops in CTR. In addition to formulating the problem, we design an online optimization algorithm that can find the optimal traffic shaping probabilities for each new user using only a cache of one number per ad contract. Experiments on a large real-world ad-serving Web portal demonstrate significant advantages over the standalone approach: a threefold reduction in underdelivery with only 10% drop in CTR, or a 2.6-fold reduction with a 4% CTR drop, and similar results over a wide range.
[1]
Deepak Agarwal,et al.
Click shaping to optimize multiple objectives
,
2011,
KDD.
[2]
Abhinandan Das,et al.
Google news personalization: scalable online collaborative filtering
,
2007,
WWW '07.
[3]
Sergei Vassilvitskii,et al.
Inventory Allocation for Online Graphical Display Advertising
,
2010,
ArXiv.
[4]
Sergei Vassilvitskii,et al.
Optimal online assignment with forecasts
,
2010,
EC '10.
[5]
Thomas P. Hayes,et al.
The adwords problem: online keyword matching with budgeted bidders under random permutations
,
2009,
EC '09.
[6]
Bee-Chung Chen,et al.
Explore/Exploit Schemes for Web Content Optimization
,
2009,
2009 Ninth IEEE International Conference on Data Mining.
[7]
Deepak Agarwal,et al.
Online Models for Content Optimization
,
2008,
NIPS.
[8]
ConitzerVincent,et al.
The ACM transactions on economics and computation
,
2013
.
[9]
John Langford,et al.
Efficient Optimal Learning for Contextual Bandits
,
2011,
UAI.