主题: Two-Sided Choices and Marketplace Management

中文题目：双边选择模型及平台市场管理

时间: 2026年1月15日 上午 9:00-10:00

地点: 管理科研楼一楼第四教室

主讲人: Dr. Yijie Zheng, Assistant Professor (Research), Department of Logistics and Maritime Studies, The Hong Kong Polytechnic University

Bio:

Yijie Zheng is an Assistant Professor (Research) in the Department of Logistics and Maritime Studies at The Hong Kong Polytechnic University. He was previously a Postdoctoral Research Fellow at the Rotman School of Management, University of Toronto. He received his Ph.D. in Decision Analytics and Operations from City University of Hong Kong in 2023 and his B.S. in Statistics from Sichuan University in 2017. His research focuses on data-driven decision-making in operations management, with an emphasis on revenue management and marketplace/platform operations.

照片：

Abstract:

Choice modeling is a critical tool for understanding consumer preferences and guiding decisions in fields such as operations management and economics. Traditional choice models typically focus exclusively on consumer decisions, overlooking the fact that, especially in online marketplaces, consumers and sellers mutually select each other. As marketplaces like Airbnb and Upwork grow in economic significance, accurately capturing these mutual choices becomes essential for effective marketplace management. In this study, we propose a two-sided feature-based choice model based on the optimal transport (OT) theory. Our model conceptualizes consumer-seller interactions as two-sided matching, where each pairing generates a surplus that incentivizes both parties to select counterparts maximizing their individual payoffs. To capture inherent randomness in consumer choices, we incorporate a bilinear stochastic surplus function within the OT framework. This formulation effectively models mutual choice behaviors and captures the complexities of contemporary marketplace dynamics. Applying this two-sided choice framework, we address critical marketplace management problems, including platform commission optimization, seller cohort size management and assortment planning, and centralized pricing decisions. Specifically, we establish the unimodality of the revenue function with respect to commission rates that enables efficient commission optimization. For seller cohort management, we demonstrate the concavity of the revenue function in relation to seller cohort sizes, and for seller assortment optimization, we develop a polynomial-time tree-decomposition algorithm. Then, for the centralized price optimization problem, we reveal that optimal centralized prices consistently feature a uniform markup. Furthermore, we provide a nonparametric learning algorithm for model estimation. Numerical results with synthetic and real-world data confirm that the proposed learning algorithm enables more accurate model estimation and yields consistent improvements in commission, assortment, and centralized pricing decisions compared with benchmark methods.