Management Science and Operations
Authors / Editors
Aouad A; Elmachtoub A N; Ferreira K J; McNellis R
We seek to provide an interpretable framework for segmenting users in a population for personalized decision-making. The standard approach is to perform market segmentation by clustering users according to similarities in their contextual features, after which a "response model" is fit to each segment to model how users respond to personalized decisions. However, this methodology is not ideal for personalization, since two users could in theory have similar features but different response behaviors. We propose a general methodology, Market Segmentation Trees (MSTs), for learning interpretable market segmentations explicitly driven by identifying differences in user response patterns. To demonstrate the versatility of our methodology, we design two new, specialized MST algorithms: (i) Choice Model Trees (CMTs) which can be used to predict a user's choice amongst multiple options, and (ii) Isotonic Regression Trees (IRTs) which can be used to solve the bid landscape forecasting problem. We provide a customizable, open-source code base for training MSTs in Python which employs several strategies for scalability, including parallel processing and warm starts. We provide a theoretical analysis of the asymptotic running time of our training method validating its computational tractability on large datasets. We assess the practical performance of MSTs on several synthetic and real world datasets, showing our method reliably finds market segmentations which accurately model response behavior. Further, when applying MSTs to historical bidding data from a leading demand-side platform (DSP), we show that MSTs consistently achieve a 5-29% improvement in bid landscape forecasting accuracy over the DSP's current model. Our findings indicate that integrating market segmentation with response modeling consistently leads to improvements in response prediction accuracy, thereby aiding personalization.