时间：2026年6月16日上午9:00-10:00

地点：财科馆317会议室

题目：RCS Approximations for Assortment Optimization with Sequential Consumer Behavior

摘要：In digital marketplaces, consumers often browse products sequentially and search within each product before making purchase decisions. Capturing such behavior can improve prediction, but it also makes assortment optimization highly complex. In practice, to preserve algorithmic speed and flexibility, platforms often simplify consumer behavior using Random Consideration Set (RCS) models, which ignore many dependencies in sequential browsing. This paper shows that such simplification is not only computationally convenient, but also theoretically justified. Specifically, we propose a Sequential Coarse-to-Fine (SCF) model that captures richer consumer behavior, nests Sequential Click-based MNL and RCS as special cases, and demonstrates improved predictive accuracy on real data. Although SCF induces a challenging optimization problem, we prove that an RCS-based assortment remains near-optimal under the full SCF model, achieving tight approximation guarantees of 0.75 when the consumer’s browsing order is known, 0.375 in the worst case when it is unknown, and 0.5 when the browsing order is uniformly random. Robustness is further established through extensions incorporating consumer impatience and generalized assortment constraints.

主讲人简介：高品，香港中文大学（深圳）数据科学学院助理教授、智能零售联合实验室副主任，曾兼任深圳市人工智能与机器人研究院副研究员。本科毕业于武汉大学物理学基地班，后赴香港科技大学攻读理论物理硕士及工业工程与决策分析博士。研究方向涵盖平台生态设计、拍卖与机制设计、离散选择建模、数据驱动决策。成果发表于Management Science、Operations Research、Manufacturing & Service Operations Management 等期刊，多次荣获全国供应链与运营管理学术年会及 POMS-China 等最佳论文奖。现主持或参与国家自然科学基金重大项目、青年科学基金（C类及B类）、滴滴盖亚青年学者基金等多项课题，并与美团、丰e足食等头部互联网企业开展深度产学研合作。