Practical Dynamic Assortment Optimization
Principal Investigator:
Professor Yu YANG
Assistant Professor, School of Data Science
Co-Principal Investigator:
Professor LI Yanzhi David
Head and Professor, Department of Marketing
Professor, Department of Management Science
Affiliated Professor, School of Data Science
Project Period: 1 November 2022 - 31 October 2024
Pathway to Impact Statement Assortment optimization aims at selecting the set of products to show to customers for the purpose of maximizing the profit of the retailer. It is one of the most important operational decisions for retailers. As customers’ preference of products are usually unknown in advance, planning the assortments shown to customers often need to trade-off between learning customers’ preference from data and optimizing the assortments using learned preference. Although existing literature has already discussed some basic online learning algorithms for dynamic assortment optimization, the problem settings are often too simple to be practical. In real retailing industry, the retailer always has many practical constraints such as inventory constraints, limited switches constraints (especially for offline retailers who cannot frequently change the assortment on the display shelf), price constraints and fairness constraints. To make dynamic assortment optimization algorithms useful in real applications, it is crucial to consider the practical and challenging constraints faced by retailers. Therefore, in this project, we plan to solve a number of important dynamic assortment planning problems under various practical constraints. We will employ the most popularly used choice model, the multinomial logit (MNL) choice model, to capture the potential customer demand and devise online algorithms with provable regret bounds. Through collaborating with online retailers such as TMall and JD.com, we seek to deploy our algorithms in real systems to solve real and practical retailing problems. This will make the assortment planning algorithms developed in this project both theoretically appealing and practically effective.
The primary objectives of this project are listed below:
- Design online algorithms with provable regret bounds for solving dynamic assortment optimization problems under various practical constraints.
- Publish 2∼4 papers in top venues such as KDD, ICML, WWW, AAAI, TKDE and JOC.
- Collaborate with online retailing platforms such as TMall and JD.com to deploy the algorithms in real systems.