EURO 2025 Leeds
Abstract Submission

1616. Residual-based Wasserstein Distributionally Robust Optimization for Assortment Planning in E-commerce Services Retail

Invited abstract in session WB-47: Empirically Driven OR in Retail, stream Retail Operations.

Wednesday, 10:30-12:00
Room: Parkinson B08

Authors (first author is the speaker)

1. Yun Long
College of Management and Economics, Tianjin University
2. Xiang Li
Meituan
3. Yanzhou Wu
Meituan
4. Yanting Jin
Meituan
5. Ning Zhu
Tianjin University
6. Jie Chen
College of Management and Economics, Tianjin University
7. Zhu Meng
8. Yani Ji
Tianjin university

Abstract

This study examines an extension of the assortment problem in E-commerce services retail. Specifically, it focuses on the challenge of deciding what types of products and what quantities to feature within a dedicated promotion section. This decision is challenged by demand uncertainty and multidimensional operational constraints. To address these challenges, this study proposes a residual-based Wasserstein Distributionally Robust Optimization (DRO) framework to model this problem. The model leverages regression analysis to incorporate side information, such as weather and holidays, into the decision-making process. By constructing data-driven ambiguity sets to quantify prediction error distribution deviations, we develop the residual-based Wasserstein DRO model that integrates forecasted demand, product quality, category coverage, and supply capacity constraints. The methodology surpasses the dependence on perfect information assumptions inherent in traditional deterministic optimization, leveraging the Wasserstein metric to characterize high-dimensional uncertainty distributions. The framework ensures robust global decision-making while preserving product quality and diversity. Numerical experiments show that the out-of-sample performance significantly surpasses baseline model, indicating enhanced decision stability in volatile market environments. The proposed approach is extensible to inventory allocation, traffic distribution, and other correlated scenarios.

Keywords

Status: accepted

Back to the list of papers