Refreshments will be served prior to the event at 3:30 p.m. in the Math Commons.
Title: Personalization in Operations and Revenue Management using Multinomial Logistic Regression: Statistical Guarantees and Insights
Abstract: Widespread interest in harnessing data and using sophisticated data analysis and prediction methods is changing the way many classical operations decision problems are being viewed. In this talk, I will explore one aspect of data-driven operations research: the customization of decisions by taking into account contextual information. I will focus on the area of revenue management, exploring the use of a customization model for pricing and assortment decisions for retailers. I show that learning under this model takes place reliably by establishing finite-sample convergence guarantees for model parameters which hold regardless of the number of possible contexts, which can be potentially uncountable. The parameter convergence guarantees are then extended to out-of-sample performance guarantees in decision problems. I will also discuss simulated experiments which demonstrate the performance of our method, as well as experiments on real transaction data for airline seating reservations which show that our method is competitive with more sophisticated and computationally intensive methods while enjoying theoretical backing that these methods do not.