Loading...

Online Dynamic Assortment Planning and Learning the Censored Demand With limited Inventory

Arhami, Omid | 2021

337 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: English
  3. Document No: 54207 (44)
  4. University: Sharif University of Technology
  5. Department: Management and Economics
  6. Advisor(s): Talebian, Masoud; Aslani, Shirin
  7. Abstract:
  8. This study considers an online multi-period assortment optimization problem over multiple replenishment cycles. The retailer chooses a subset from N substitutable products and decides how many of each product to order and sell at each time period. Retailer is constrained by a total inventory capacity, a cardinality constraint on the product variety (display space), and predetermined replenishment time intervals. The assortment selection is modeled as a Multi-armed Bandit problem and the customers' choice is modeled by the Multinomial Logit (MNL) choice model with dynamic substitution. The objective is to optimize the revenue by actively learning the censored demand and improving the offering composition at every period. In this novel approach, the offering and consequently the exploration-exploitation decision has two dimensions: the assortment members and the inventory allocation quota. To circumvent the censored demand effect, our algorithm detects and filters out the curtailed demand data in an implementable way. In this study we develop a dynamic assortment optimization algorithm that demonstrates numerical performance results on a par with the best algorithms in the literature when capacity is larger than a minimum. We show the dependence of the order of regret on inventory constraint and observe that this constraint has a significant impact on both the learning and the profit of a retailer who tries to learn the demand and the best inventory composition on the fly. We also show that despite what previously has been proven in the literature for the full-information setting, choosing the minimum variety in the assortment leads to better results in the case of very tight inventory constraint
  9. Keywords:
  10. Simulation ; Demand Learning ; Multi-Armed Bandit Problem ; Assortment Planning ; Planning Inventory ; Customer Choice Behavior ; Thompson Sampling

 Digital Object List

 Bookmark

No TOC