Large-Scale Personalized Delivery for Guaranteed Display Advertising with Real-Time Pacing
Zhen Fang, Yang Li, Chuanren Liu, Wenxiang Zhu, Yu Zheng, Wenjun Zhou · 2019
Guaranteed display (GD) has been a successful model for display advertising. Existing solutions usually model GD services as a crowd-level supply allocation problem. This formulation, however, not only ignores user heterogeneity within crowds, but also makes it difficult to incorporate individual-level constraints. In this paper, we present an large-scale system for personalized delivery in GD advertising services. A unique contribution is to model the allocation problem at the individual level that accounts for user-ad interactions. Therefore, our system can conveniently incorporate complex constraints, such as the priority of GD contracts, the display frequency of ads, and the effectiveness of ad slots arrangement. Moreover, we develop a real-time pacing strategy to fulfill GD contracts with smooth ad delivery and optimized ad performance, such as cost-per-click (CPC) and cost-per-action (CPA). Our system can be parallelized to efficiently compute the delivery solution with billions decision variables. Using both offline evaluation and online A/B tests, we demonstrate that our solution is effective in terms of both accuracy and efficiency.