数据驱动下定向展示广告保量投放鲁棒策略

Data-driven robust strategy for guaranteed delivery of targeted display advertising

具有不确定性的曝光供应量给广告资源的合理分配带来了极大挑战.针对这种不确定性,本文提出了数据驱动分布鲁棒定向展示广告保量投放分配模型.首先考虑以最大化发布商收益为目标,构建对未满足需求或超出需求的部分均进行惩罚的带有机会约束的随机规划模型.其次,利用曝光供应量的历史数据,定义Wasserstein不确定集,构建数据驱动分布鲁棒机会约束模型,该模型提出的投放策略确保在曝光供应量最坏分布情况下发布商收益最大.通过保守近似,模型可重构为易于求解的混合整数规划问题.最后,大规模样本外测试验证了模型和求解方法的可行性、有效性以及稳定性.

The uncertainty in impression supply presents a significant challenge to the optimal allocation of advertising resources.To address this uncertainty,this paper proposes a data-driven distributionally robust model for targeted display ad allocation problem.Firstly,a stochastic programming model with chance constraints is formulated,with the objective of maximizing the publisher’s revenue and penalizing both the unmet demand and the excess of demand.Second,using historical impression supply data,a data-driven distributionally robust chance-constrained model is established.This model utilizes the Wasserstein ambiguity set to propose an allocation strategy that maximizes the publisher’s revenue even under the worst-case distribution of impression supply.Through a conservative approximation,the model can be reformulated as an easy-to-solve mixed-integer programming problem.Finally,large-scale out-of-sample experiments are conducted to validate the feasibility,efficiently,and stability of the model and the solving approach.