Allocation of Overlapping Inventory in Online Advertising via MultiObjective Optimization

We discuss the allocation of inventory, defined by pools of users possessing multiple characteristics, e.g. age-range, ge- ographical location, etc., to on-line advertisers who may seek some or all of these characteristics. Since the pools may overlap, some form of optimization model is called for. We first consider a linear, i.e. LP model, which maximizes short term revenue and then several non-linear variants which tend to optimize advertiser “fairness”. We then consider multi-objective models which attempt to balance these two objectives and suggest a computational procedure, illustrated by a toy example as well as real data from a display advertising system.