Premium
A Planning Approach to Revenue Management for Non‐Guaranteed Targeted Display Advertising
Author(s) -
Shen Huaxiao,
Li Yanzhi,
Guan Jingjing,
Tso Geoffrey K.F.
Publication year - 2021
Publication title -
production and operations management
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 3.279
H-Index - 110
eISSN - 1937-5956
pISSN - 1059-1478
DOI - 10.1111/poms.13275
Subject(s) - computer science , revenue , display advertising , common value auction , budget constraint , online advertising , poisson regression , set (abstract data type) , plan (archaeology) , operations research , resource allocation , poisson distribution , point (geometry) , integer programming , mathematical optimization , integer (computer science) , revenue management , linear programming , event (particle physics) , resource (disambiguation) , economics , microeconomics , the internet , algorithm , mathematics , population , history , computer network , archaeology , sociology , world wide web , geometry , accounting , quantum mechanics , programming language , statistics , physics , demography
Many publishers of online display advertising sell their ad resources through event‐based auctions in the spot market. Such a way of selling lacks a holistic view of the publisher’s ad resource and thus suffers from a well‐recognized drawback: the publisher’s revenue is often not maximized, particularly due to users’ dynamic ad clicking behavior and advertisers’ budget constraints. In this study, we propose a planning approach for ad publishers to better allocate their ad resources. Specifically, we propose a framework comprising two building blocks: (i) a mixed‐integer nonlinear programming model that solves for the optimal ad resource allocation plan, which maximizes the publisher’s revenue, for which we have developed an efficient solution algorithm; and (ii) an arbitrary‐point‐inflated Poisson regression model that deals with users’ ad clicking behavior, whereby we directly forecast the number of clicks, instead of relying on the click‐through rate (CTR) as in the literature. The two blocks are closely related in the sense that the output of the regression model serves as the input to the optimization model and the optimization model motivates the development of the regression model. We conduct extensive numerical experiments based on a data set spanning 20 days provided by a leading social network sites firm. Experimental results substantiate the effectiveness of our approach.