Speaker: Ezgi Karabulut Türkseven (Sabancı University)
Title: Bundle Pricing via Learning the Market from Customer Preferences
Date & Time: 16 October 2026, 13:30
Place: EA 302
Abstract: We study how a seller offering n items through mixed bundling can learn the structure of a multi-segmented market from customer purchase choices alone, without observing valuations or segment membership. Because all 2^n bundles must be priced, generic black-box optimization has to search an O(2^n)-dimensional space, which becomes inefficient as n grows. We propose BLMP (Bundle Pricing via Learning the Market from Customer Preferences), which decomposes bundle-level preference feedback into item-specific data, so that item valuations and segment probabilities are learned in O(n) dimensions. These estimates are then used in a mixed-integer program to compute near-optimal bundle prices. Against CMA-ES and GP-UCB, BLMP’s final solutions stay within 1% of the full-information benchmark under deterministic valuations and within 5% under probabilistic valuations. CMA-ES performs poorly, and GP-UCB’s runtime grows severely with the number of interactions.
Bio: Ezgi Karabulut Türkseven is an assistant professor in the Industrial Engineering program at Sabancı University. She holds BSc and MSc degrees from the Industrial Engineering department at Boğaziçi University and received her PhD in Operations Research from the Georgia Institute of Technology. Her research lies at the intersection of operations research, machine learning, and optimization, with a particular focus on integrating preference learning into decision-making models. Her recent work explores the use of data-driven approaches in multi-period optimization problems with multiple agents.