{"title":"Dynamic allocation of display advertising impressions in dual sales channels","authors":"","doi":"10.1016/j.omega.2024.103213","DOIUrl":null,"url":null,"abstract":"<div><div>We study a multi-period ad allocation problem faced by an online publisher who sells ad impressions on websites through two sales channels. In the guaranteed sales channel, advertisers submit heterogeneous offers for contracts under which the publisher guarantees delivery of a certain number of ad impressions over a certain period; in the real-time bidding (RTB) sales channel, the publisher runs an RTB auction to sell ad impressions. In each period, the publisher decides whether to accept or reject contract proposals; how to allocate ad impressions across existing contracts; and how many impressions to sell via RTB. The publisher faces uncertain demand from advertisers and an uncertain supply of impressions, which are generated by viewers visiting the publisher’s websites. We formulate the problem as a finite-horizon stochastic dynamic program, which poses significant methodological challenges. We first present structural properties of optimal policies under certain cases. To avoid the curse of dimensionality in dynamic programming, we develop an approach involving Lagrangian relaxations. We decompose the problem into a series of solvable subproblems and derive optimal policies. We further develop Lagrangian policies with performance guarantees. We show that when Lagrange multipliers depend on more signal history, the linear term’s weight of the number of contract types in the performance upper bound decreases. Furthermore, if the Lagrange multipliers depend on the full signal history, the corresponding Lagrangian policies will be asymptotically optimal to the number of contract types. We also explore a more suitable case for large-scale real-time ad allocation and create Lagrangian policies that yield comparable performance guarantees. Finally, we extend our main results to four new scenarios.</div></div>","PeriodicalId":19529,"journal":{"name":"Omega-international Journal of Management Science","volume":null,"pages":null},"PeriodicalIF":6.7000,"publicationDate":"2024-10-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Omega-international Journal of Management Science","FirstCategoryId":"91","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0305048324001774","RegionNum":2,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"MANAGEMENT","Score":null,"Total":0}
引用次数: 0
Abstract
We study a multi-period ad allocation problem faced by an online publisher who sells ad impressions on websites through two sales channels. In the guaranteed sales channel, advertisers submit heterogeneous offers for contracts under which the publisher guarantees delivery of a certain number of ad impressions over a certain period; in the real-time bidding (RTB) sales channel, the publisher runs an RTB auction to sell ad impressions. In each period, the publisher decides whether to accept or reject contract proposals; how to allocate ad impressions across existing contracts; and how many impressions to sell via RTB. The publisher faces uncertain demand from advertisers and an uncertain supply of impressions, which are generated by viewers visiting the publisher’s websites. We formulate the problem as a finite-horizon stochastic dynamic program, which poses significant methodological challenges. We first present structural properties of optimal policies under certain cases. To avoid the curse of dimensionality in dynamic programming, we develop an approach involving Lagrangian relaxations. We decompose the problem into a series of solvable subproblems and derive optimal policies. We further develop Lagrangian policies with performance guarantees. We show that when Lagrange multipliers depend on more signal history, the linear term’s weight of the number of contract types in the performance upper bound decreases. Furthermore, if the Lagrange multipliers depend on the full signal history, the corresponding Lagrangian policies will be asymptotically optimal to the number of contract types. We also explore a more suitable case for large-scale real-time ad allocation and create Lagrangian policies that yield comparable performance guarantees. Finally, we extend our main results to four new scenarios.
期刊介绍:
Omega reports on developments in management, including the latest research results and applications. Original contributions and review articles describe the state of the art in specific fields or functions of management, while there are shorter critical assessments of particular management techniques. Other features of the journal are the "Memoranda" section for short communications and "Feedback", a correspondence column. Omega is both stimulating reading and an important source for practising managers, specialists in management services, operational research workers and management scientists, management consultants, academics, students and research personnel throughout the world. The material published is of high quality and relevance, written in a manner which makes it accessible to all of this wide-ranging readership. Preference will be given to papers with implications to the practice of management. Submissions of purely theoretical papers are discouraged. The review of material for publication in the journal reflects this aim.