Data Adapted Consumer Choice Modeling

Abstract:

In this talk, we will discuss how one may without fixing a specific consumer choice model, such as the Multinomial, Exponomial or Markovian Chain, consider a general choice model that will adapt to the size and breadth of the data currently available and that may arrive in the future.

We do this by developing a general one-to-one representation of any customer choice probability model by a token tree. To do this, we will focus on characterizing the random choices of a representative customer as a random set function. The support of this random set function is observed to be the vertices of a polytope that is the product space of Simplex.

Monotonicity or partial monotonicity of this set function will lead to a collection of models progressing from rational (temporal tree) model to completely irrational (token tree) consumer choice models. We will discuss how this token tree and operational data analytics (ODA) can be used to adaptively identify appropriate model for the data at hand.

Speaker:

Professor J. George Shanthikumar
Richard E. Dauch Distinguished Chair in Manufacturing and Operations Mgmt
Supply Chain and Operations Management
Mitch Daniels School of Business

Purdue University