2016 INFORMS Annual Meeting Program

MD66

INFORMS Nashville – 2016

2 - Assigning Students To Schools To Minimize Socioeconomic Variation Between Schools Richard Forrester, Associate Professor of Mathematics, Dickinson College, College and Louther Streets, Carlisle, PA, 17013, United States, forrestr@dickinson.edu, Kevin Hutson, Elizabeth Bouzarth Numerous studies have found that a student’s academic achievement is as much determined by the socioeconomic composition of their school as their own socioeconomic status. In this talk we provide a methodology for assigning students to schools so as to balance the socioeconomic compositions of the schools while taking into consideration the total travel distance. Our technique utilizes a bi-objective general 0-1 fractional program that is linearized into a mixed 0-1 linear program which can be submitted directly to a standard optimization package. As a test case for our approach we analyze data from the Greenville County School District in Greenville, South Carolina. 3 - Consumer's Preferences Modeling For Rail Transportation In Qatar Rana Sobh, Qatar University, Doha, Qatar; r.sobh@qu.edu.qa, Belaid Aouni The increase in traffic congestion, road safety and pollution have led Qatar to improve the existing public transportation system and introduce Doha Metro. This shift in public transportation requires changes in consumers’ perceptions about rail transportation. The aim of our paper is to predict the factors that may impact consumers’ behavior and their preferences in choosing transportation sys- tems. Moreover, our study aims to develop a better understanding of the rail transportation mode adoption in Qatar and provides some recommendations to the policy makers in Qatar Rail. MD65 Mockingbird 1- Omni Data Analytics and Machine Learning Sponsored: Information Systems Sponsored Session Chair: Sriram Somanchi, University of Notre Dame, 344 Mendoza College of Business, Notre Dame, IN, 46556, United States, somanchi.1@nd.edu 1 - Analytical And Empirical Modeling Of Complementarities In An Online Advertising Supply Chain Changseung Yoo, The University of Texas at Austin, Austin, TX, United States, csyoo@utexas.edu, Anitesh Barua, Genaro Gutierrez We examine channel structures and pricing models in an online advertising supply chain using a proprietary dataset. We develop analytic as well as structural econometric models that enable us to model interactions between the channel structures/pricing models, and quantify synergy effects between them. To the best of our knowledge, our study takes the first step of analyzing details of an online advertising supply chain. Moreover, while the extant literature emphasizes choosing between pricing models, we demonstrate that using multiple models in concert yields higher overall profitability due to spillover effects and strategic complementarities among the pricing schemes. 2 - Predicting Hotel Revenue Using Hotel Latent Quality Uttara Madurai Ananthakrishnan, Carnegie Mellon University, umadurai@andrew.cmu.edu Online reviews have become a major source of information for consumers in the past decade and influence various aspects of e-commerce such as purchase decisions and product sales in a variety of settings. Most of the work in this field has focused on numerical ratings to understand the impact of online reviews on sales. In our paper, we use a novel topic-modeling technique on a large dataset of online reviews of hotels and study if topics obtained from this technique provide a better representation of a hotel’s quality. We then analyze how each hotel’s quality evolves over time and predict the changes in hotel revenue using such topics. 3 - SkillR: Personalized Skill Recommendations Using Joint Bayesian Member-job Clustering Abhinav Maurya, Carnegie Mellon University, Pittsburgh, PA, United States, ahmaurya@cmu.edu, Rahul Telang, Sai Sundar Skill gaps in various sectors of the economy are considered to be major problems facing economies today. Increasing the productivity of a member of the workforce depends on recommending skills whose acquisition will yield the highest utility gains for the member. We present SkillR - a skill recommendation algorithm that employs a joint Bayesian clustering model to match members to other similar members as well as relevant jobs, and to identify the top skills that provide the maximum utility gains to a member. Our evaluation suggests that SkillR leads to orders of magnitude improvement in the job propensity of recommended skills compared to a traditional collaborative filtering system.

4 - Does Government Surveillance Give Twitter The Chills? Sriram Somanchi, University of Notre Dame, somanchi.1@nd.edu, Laura Brandimarte, Edward McFowland, Uttara Madurai Ananthakrishnan Since Snowden’s revelations regarding mass surveillance programs implemented by the NSA, Government surveillance has garnered huge attention. The research community has attempted to estimate the “chilling effects” of surveillance, the tendency to self-censor. Until now, such effects have been estimated using either the search terms, Wikipedia articles, or survey data. In this work, we propose a new method in order to test for chilling effects in online social media. We use large Twitter dataset and propose the use of new statistical machine learning method in order to detect anomalous trends in user behavior after Snowden’s revelations made users aware of existing surveillance programs. MD66 Mockingbird 2- Omni Computer Experiments and Uncertainty Quantification Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Ying Hung, Rutgers State University of New Jersey, Piscataway, NJ, United States, yhung@stat.rutgers.edu 1 - Robust Parameter Design Using Computer Experiments Roshan Vengazhiyil, Georgia Institute of Technology, roshan@gatech.edu Space-filling designs, commonly used in computer experiments, try to spread out points uniformly in the experimental region. However, when the objective is to achieve robustness against noise factors, uniformity is no longer needed in the space of noise factors. This is because noise factors usually follow nonuniform distributions such as normal distribution. It makes more sense to place points in the high probability regions where more “actions” take place. Unfortunately, nonuniform points in the experimental region can lead to problems in model fitting. In this article we propose novel design and modeling strategies to deal with these issues. 2 - Invariance-preserving Emulation For Computer Models, Simulation models with invariance properties appear in material science, physics, biology and other fields. Standard emulation methods such as Gaussian process regression cannot accommodate input invariance and thus do not work for this new problem. We will propose a kernel-based emulation method to preserve invariance in inputs. The method employs a direct graph representation and equivalence relations to characterize relabeling invariance. The effectiveness of the proposed method is illustrated by using several examples from material science. 3 - A Sequential Maximum Projection Design Framework For Computer Experiments With Inert Factors Shan Ba, The Procter & Gamble Company, Cincinnati, OH, United States, ba.s@pg.com Many computer experiments involve a large number of input factors, but many of them are inert and only a subset are important. In this talk we present a new sequential design framework which can accommodate multiple responses and quickly screen out inert factors so that the final space-filling design is close to optimal with respect to the active factors. The new approach does not require prescribing the total sample size, and under the presence of inert factors, it can lead to substantial savings in simulation resources. Even if all the factors are important, the proposed sequential design can still achieve similar overall space- filling property compared to a maximin LHD optimized in a single stage. With Application To Structural Energy Prediction Peter Qian, University of Wisconsin - Madison, peter.qian@wisc.edu

229