Probabilistic simulation of spatial demand for intelligent product allocation

Connecting consumers with relevant products is a very important problem in both online and offline commerce. In many offline retail settings, product distributors extend bids to place and manage product displays within a retail outlet. The distributor aims to choose a spatial allocation strategy that maximizes revenue given a preset budget constraint. Prior work shows that carefully selecting product locations within a store can minimize search costs and and induce consumers to make "impulse"purchases. Such impulse purchases are influenced by the spatial configuration of the store. However, learning important spatial patterns in offline retail is challenging due to the scarcity of data and the high cost of exploration and experimentation in the physical world. To address these challenges, we propose a stochastic model of spatial demand in physical retail, which we call the Probabilistic Spatial Demand Simulator (PSD-sim). PSD-sim is an effective mirror of the real environment because it exploits the structure of common retail datasets through a hierarchical parameter sharing structure, and is able to incorporate spatial and economic knowledge through informative priors. We show that PSD-sim can both recover ground truth test data better than baselines, and generate new data for unseen states. The simulator can naturally be used to train policy estimators that discover intelligent, spatial allocation strategies. Finally, we perform a preliminary study into different optimization techniques and find that Deep Q-Learning can learn an effective spatial allocation policy.