Abstract
Retailing and media platforms recommend two types of items to their users: sponsored items that generate ad revenue and nonsponsored ones that do not. The platform selects sponsored items to maximize ad revenue, often through programmatic auctions, and nonsponsored items to maximize user utility with a recommender system (RS). We develop a binary integer programming model to allocate sponsored recommendations considering dual objectives of maximizing ad revenue and user utility. We propose an algorithm to solve it in a computationally efficient way. Our method is a form of postfiltering to a traditional RS, making it widely applicable in two-sided markets. We apply the algorithm to data from an online grocery retailer and show that user utility for the recommended items can be improved while reducing ad revenue by a small amount. This multiobjective approach unifies programmatic advertising and RS and opens a new frontier for advertising and RS research. We provide an extended discussion of future research topics.
ACKNOWLEDGMENTS
We thank Özge Sürer for helpful discussions and Northwestern University's Spiegel Research Center for supporting the second and third authors.
Notes
1 We do not allow because then the algorithm would have no way of allocating nonsponsored items when, for example,
. When
is very small, e.g.,
, then up to
sponsored items are selected based on the ad revenue, and the remaining nonsponsored items, which have ad revenue equal to 0 causing the ad revenue term in (2) to equal 0, are selected based on user utility. Constraining
also addresses other degenerate cases, such as not having enough sponsored items to fill
slots.
2 Sometimes it is more convenient to express this constraint as an upper bound on the fraction of sponsored items rather than the number. In such cases, one can compute by multiplying the upper-bound fraction by
and using the algorithm as stated.
3 Cosine similarity is one of the most commonly used measures of the similarity between two users. If we think of each user’s ratings as n-vector, it is the cosine of the angle between them. When user ratings are centered to have mean 0 for each user, cosine similarity is the Pearson correlation between the row vectors. Values near 1 indicate highly similar users, while values near 0 mean low similarity.
Additional information
Notes on contributors
Edward C. Malthouse
Edward C. Malthouse (PhD, Northwestern University) is the Erastus Otis Haven professor and research director of the Spiegel Research Center, Medill School of Journalism, Media, Integrated Marketing Communications, Northwestern University.
Yasaman Kamyab Hessary
Yasaman Kamyab Hessary (PhD, University of North Carolina in Charlotte) is a postdoctoral fellow at the Spiegel Research Center, Medill School of Journalism, Media, Integrated Marketing Communications, Northwestern University.
Khadija Ali Vakeel
Khadija Ali Vakeel (PhD, Indian Institute of Management, Indore) is a postdoctoral fellow at the Spiegel Research Center, Medill School of Journalism, Media, Integrated Marketing Communications, Northwestern University.
Robin Burke
Robin Burke (PhD, Northwestern University) is a professor in the Department of Information Science, College of Media, Communication, and Information, University of Colorado Boulder.
Morana Fudurić
Morana Fudurić (PhD, Università della Svizzera italiana) is an assistant professor in the Department of Marketing, Faculty of Economics and Business, University of Zagreb.