Abstract
Uplift modeling is aimed at estimating the incremental impact of an action onan individual's behavior, which is useful in various application domains suchas targeted marketing (advertisement campaigns) and personalized medicine(medical treatments). Conventional methods of uplift modeling require everyinstance to be jointly equipped with two types of labels: the taken action andits outcome. However, obtaining two labels for each instance at the same timeis difficult or expensive in many real-world problems. In this paper, wepropose a novel method of uplift modeling that is applicable to a morepractical setting where only one type of labels is available for each instance.We show a generalization error bound for the proposed method and demonstrateits effectiveness through experiments.