2015 Informs Annual Meeting

WC11

INFORMS Philadelphia – 2015

2 - Clickstream Big Data and “Delivery Before Order Making” for Online Retailers Haoxuan Xu, School of Management, Huazhong University of Science and Technology, 1037 Luoyu Road, Hongshan District, Wuhan, 430074, China, juwan.hsu@gmail.com, Yeming Gong, Wilco Van Den Heuvel, Albert Wagelmans, Jinlong Zhang Our research is inspired by a leading online retailer using clickstream big data to estimate customer demand and then ship items to customers by a mode of “Delivery Before Order Making” (DBOM) operational mode. Using clickstream data to obtain advance demand information (ADI) in order quantities, we integrate the forecasting with a single-item uncapacitated dynamic lot sizing problem in a rolling-horizon environment. Using the simulated clickstream data, we evaluate the performance of DBOM mode. 3 - Estimating Seasonality of E-commerce Sales Abhay Jha, Walmart E-Commerce, 850 Cherry, San Bruno, CA, United States of America, abhaykj@gmail.com The assortment of items in e-commerce includes a lot of items with short life- span; hence the traditional methods of estimating seasonality per item by looking at past years’ sales are not applicable here. We will formulate this problem as maximizing the penalized likelihood of a state space model, where we penalize the seasonality to have some plausible properties, and use the semantic information about items to constrain similar items to have similar seasonality. 4 - A Simulation Framework of Consumer-to-Consumer Ecommerce Business Model Oloruntomi Joledo, University of Central Florida, 4000 Central Florida Blvd, Orlando, FL, 32816, United States of America, Tomi.Joledo@knights.ucf.edu, Luis Rabelo In the past decade, ecommerce transformed the business models of many companies.This paper proposes a modeling and simulation framework to investigate how the actions of stakeholders in consumer-to-consumer ecommerce affect the system performance as well as the business dynamics of the model. The goal is to provide stakeholders with a decision making tool to assess the viability and performance of the consumer-to-consumer business model. 5 - Higher Prices for Larger Quantities? Non-monotonic Price-quantity Relations in B2b Markets Wei Zhang, Assistant Professor, University of Hong Kong, University of Hong Kong, Hong Kong, China, zhangw.03@gmail.com We study a microprocessor company that has a limited capacity and negotiates with each buyer for the price. Our analysis of their data reveals that larger purchases do not always result in bigger discounts, and we show that the non- monotonicity is rooted in how sellers value capacity. The value of residual capacity may be initially convex and then concave. Such a value function is sufficient to ensure a non-monotonic price-quantity relationship. Chair: Ioannis Fragkos, Post Doctoral Fellow, HEC Montreal, 3000 Chemin de la Cote-Sainte-Catherine, Montreal, Canada, ioannis.fragkos@cirrelt.ca 1 - A Computational Study of Two-Period Relaxations for Big-Bucket Lot-Sizing Problems Ioannis Fragkos, Post Doctoral Fellow, HEC Montreal, 3000 Chemin de la Cote-Sainte-Catherine, Montreal, Canada, ioannis.fragkos@cirrelt.ca, Mahdi Doostmohammadi, Kerem Akartunali Lot-sizing problems form the backbone of most modern production planning systems. Despite the significant advancements in optimization theory and software, most methods used in practice lead to higher-than-optimal costs. In this talk we investigate new classes of inequalities that are based on two-period relaxations. We discuss separation procedures and a branch-and-cut implementation. Computational experiments are promising, and show that the proposed inequalities derive improved lower bounds. 2 - A Small-Order-Polynomial-Sized Linear Program for the Traveling Salesman Problem with Tight Bounds WC11 11-Franklin 1, Marriott Optimization Integer Programming II Contributed Session

3 - A Branch and Bound Approach to the Minimum K-enclosing Ball Problem Marta Cavaleiro, Rutgers University, 100 Rockefeller Rd., Piscataway, NJ, 08854, United States of America, marta.cavaleiro@rutgers.edu, Farid Alizadeh The minimum k-enclosing ball problem seeks the ball with smallest radius that contains at least k of n given points. This problem is NP-hard. For the minimum enclosing ball problem (requiring the ball to contain all points) there are both primal and dual iterative algorithms that are very similar to the simplex method for LP. We incorporate these methods into a branch and bound search to solve the minimum k-enclosing ball problem. Some computational results will be presented. 4 - Cutting Circles via Piecewise Milp Steffen Rebennack, Colorado School of Mines, 1500 Illinois Street, Golden, CO, United States of America, srebenna@mines.edu In circle cutting, one computes an area minimizing rectangle hosting a given set of circles. These circles are not allowed to overlap. This circle cutting problem is typically formulated as a continuous NLP problem where the aforementioned nonoverlap condition make its feasible region nonconvex. We approximate these nonoverlap conditions and the bilinear objective function with piecewise linear and tailored constructs. In doing so, the resulting formulation becomes a MILP problem. Chair: Michael Metel, PhD Student, McMaster University, 1280 Main St. West, Hamilton, ON, L8S4M4, Canada, Michael Metel 1 - A Bilevel Programming Model: Reduction of Dimension of the Upper Level Problem Vyacheslav Kalashnikov, Assist. Prof., Tecnologico de Monterrey (ITESM), Campus Monterrey, 2501 Av. Eugenio Garza Sada South, Monterrey, NL, 64849, Mexico, kalash@itesm.mx, Nataliya Kalashnykova Bilevel stochastic programming is often applied to model interaction between a Natural Gas Shipping Company and a Pipeline Operating Company. The problem is reduced to an also bilevel model but with linear constraints. However, this reduction makes the dimension of the upper level problem an unbearable burden even for the modern PC systems. The aim of this paper is a mathematical formalization of the reduction of the upper level problem’s dimension without affecting the optimal solution. 2 - Optimization Problem with a Reference Utility Based Stochastic Dominance Constraint Jian Hu, Assistant Professor, University of Michigan- Dearborn, 2340 Engineering Complex, 4901 Evergreen Rd, Dearborn, MI, 48128, United States of America, jianhu@umich.edu, Gevorg Stepanyan We address a novel approach to relax the second order stochastic dominance, which characterizes a norm-based functional perturbation region based on a reference utility function recommended by the decision maker. This approach best represents the decision maker’s individual preference. We discuss an optimization problem using this dominance constraint, and provide a solution method using Bernstein polynomial approximation. 3 - Hybrid Robust-stochastic Optimization Approach for Closed-loop Supply Chain Network Design Esmaeil Keyvanshokooh, PhD Student, University of Michigan, 1205 Beal Ave., Ann Arbor, MI 48109-2117, Ann Arbor, MI, United States of America, keyvan@umich.edu, Elnaz Kabir, Sarah Ryan Our contribution is to develop a novel hybrid robust-stochastic programming approach to simultaneously model two different types of uncertainties by including stochastic scenarios for transportation costs and polyhedral uncertainty sets for demands and returns. An accelerated stochastic Benders decomposition is proposed for solving this model. Numerical studies are performed to show the benefits of our approach. WC12 12-Franklin 2, Marriott Optimization Stochastic III Contributed Session

Mark Karwan, University at Buffalo, 342 Bell Hall, North Campus, Buffalo, NY, 14260, United States of America, mkarwan@buffalo.edu, Moustapha Diaby, Lei Sun

We present a polynomial-sized linear program for the n city TSP drawing upon ‘complex flow’ modeling ideas by the authors who used an O(n9)xO(n8) model. Here we have only O(n5) variables and O(n4)constraints. We use an assignment problem-based abstraction of tours not employing the traditional city-to-city variables of the standard TSP formulation. We solved thousands of problems with up to 26 cities using the simplex and barrier methods of CPLEX, consistently obtaining all integer solutions.

430