2015 Informs Annual Meeting

MD18

INFORMS Philadelphia – 2015

MD16 16-Franklin 6, Marriott Application of Linear and Conic Programs with Complementarity Constraints Sponsor: Optimization/Linear and Conic Optimization Sponsored Session Chair: Xin Shen, RPI, 110 8th Street, Troy, NY, 12180, United States of America, shenx5@rpi.edu 1 - Robust Optimization for Network Design Problems with Equilibrium Flows Liu Su, Iowa State University, 0076 Black Engineering, Ames, IA, 50011, United States of America, suliu@iastate.edu, Lizhi Wang, Guiping Hu To identify optimal network capacity expansion, we build up a bi-level model for network design problems with equilibrium flows under the robust optimization paradigm. We transformed the lower level problem into a mixed-integer linear program and use a branch and cut algorithm to solve the bi-level optimization problem. 2 - Application of Complementarity Problems in Multibody Dynamics and Robotics Ying Lu, Rensselaer Polytechnic Institute, 2408 21st St. Apt. 6, Troy, NY, 12180-1811, United States of America, rosebudflyaway@gmail.com, Jeff Trinkle Frictional contacts in multibody dynamics and robotic simulation are generally written as differential Complementarity Problems (dCPs), which are solved as a series of CPs. We compare several models and solution algorithms, as well as a GPU based CUDA parallel solver to to solve the Complementarity Problems arising in physical simulation. 3 - Property of a Relaxation Scheme for Rank Constrained Optimization Problems Xin Shen, RPI, 110 8th Street, Troy, NY, 12180, United States of America, shenx5@rpi.edu, John Mitchell Recently rank constrained optimization problems have received increasing interest because of their wide application. This class of problems has been considered computationally challenging because of its nonconvex nature. In this talk we focus on a mathematical program with semidefinite cone complementarity constraints formulation of the class. We’ll consider a relaxation scheme for the formulation and discuss its properties including stationary conditions and local optimality. 4 - Heuristics for QPLCCS using Nlp Solvers Aided by Semidefinite Relaxations Patricia Gillett, PhD Candidate, Département de Mathématiques

2 - Impact of Sub-networks on the Diffusion of Innovation Xu Dong, Research Assistant, University of Miami, 1251 Memorial Drive, Coral Gables, FL, 33146, United States of America, x.dong3@umiami.edu, Nazrul Shaikh Extant research shows that the structural properties of social networks influence the diffusion of innovation; however, these studies assume that the network is one giant cluster. Networks can have disconnected clusters (sub-networks) that introduce discontinuities in the diffusion pathways. Our research provides an understanding of the impact of discontinuities on diffusion. 3 - Identifying High Value Customers in a Network: Individual Characteristics Versus Social Influence Sang-Uk Jung, Assistant Professor, Hankuk University of Foreign Studies, Imunro 102, Dongdaemun-gu, Seoul, 130-791, Korea, Republic of, sanguk.jung@hufs.ac.kr, Qin Zhang, Gary Russell Firms are interested in identifying customers who generate the highest revenues. In a social network setting, customer interactions can play an important role in purchase behavior. This study proposes a spatial autoregressive model that explicitly shows how network effects and individual characteristics interact in generating firm revenue.Using model output, we develop a method of identifying individuals whose purchase behavior most impacts the total revenues in the network. Chair: Onur Seref, Virginia Tech, 2060 Pamplin Hall (0235), Blacksburg, VA, 24061, United States of America, seref@vt.edu 1 - A Tangled Web: Evaluating the Impact of Displaying Fraudulent Reviews on Review Portals Uttara Ananthakrishnan, Carnegie Mellon University, 5000 Forbes Ave, Pittsburgh PA 15213, United States of America, umadurai@andrew.cmu.edu, Michael D Smith, Beibei Li This paper studies how users respond to fraudulent reviews and how platforms can leverage such knowledge to design better fraud management. We combine randomized experiments, behavioral economics with machine learning using large-scale data from Yelp. We find to improve user trust platforms should display the fraudulent information. Finally, our statistical analysis using MLE allows us to design a novel fraud-awareness reputation system. 2 - Strength in Numbers: Can Big Data Eliminate the Need for Complex Opinion-mining Algorithms? Opinion mining, defined as the task of identifying the polarity of text segments, is a building block for many popular applications. Relevant methods have evolved from simple lexicon-based approaches to expensive NLP algorithms that try to emulate human thought. What if there was a simpler way to capture complex linguistic patterns, by utilizing the unprecedented availability of opinion-rich datasets? Our work addresses this question and motivates a new line of applications for Big Data. 3 - Restaurant Hygiene Grades and Online Reviews Jorge Mejia, University of Maryland, Robert H. Smith School of We focus on understanding the relationship between online reviews and a significant public health problem: restaurant-related foodborne illness. Recent initiatives to publicize the results of restaurant health inspections have been shown to reduce the occurrence of foodborne illness. We use the semantic information in online reviews to forecast health inspection results for restaurants in NYC. This approach can be used to improve the effectiveness of health inspection programs. 4 - Automatic Sequence Extraction for Sequence Alignment in Text Mining Michelle Seref, Virginia Tech, Pamplin 1007, 0235, Blacksburg, VA, 24061, United States of America, mmhseref@vt.edu, Onur Seref We illustrate novel methods to automatically extract sequences from pre-labeled text in order to apply sequence alignment for classifying text. Sequences are initially generated using n-gram approaches and then aggregated into semantically unique sequences. Sequence alignment uses these sequences to detect semantically equivalent text with either exact word or synonym matches. We demonstrate our method on several text domains. Theodoros Lappas, Assistant Professor, Stevens Institute of Technology, 335 Washington, Apt. 2, Hoboken, NJ, 07030, Greece, tedlappas@gmail.com Business, College Park, MD, United States of America, jmejia@rhsmith.umd.edu, Shawn Mankad, Anand Gopal MD18 18-Franklin 8, Marriott Methodologies in Text Mining for Big Data Cluster: Modeling and Methodologies in Big Data Invited Session

et de Génie Industriel, École Polytechnique de Montréal, Montréal, QC, Canada, patricia-lynn.gillett@polymtl.ca, Miguel Anjos, Joaquim Júdice

We present a semidefinite programming relaxation technique with iterative cutting planes for quadratic programs with linear complementarity constraints (QPLCC). We discuss how an optimal solution to the SDP relaxation can be used to warmstart the solution of the QPLCC using common local and global NLP solvers. We report some numerical results demonstrating the quality of the SDP bound and the effectiveness of the warmstarting procedures.

MD17 17-Franklin 7, Marriott Modeling Social Influence in Networks Sponsor: Optimization/Network Optimization Sponsored Session

Chair: Vladimir Boginski, University of Florida, 303 Weil Hall, Gainesville, FL, United States of America, boginski@reef.ufl.edu 1 - Fashion Supply Chain Network Competition with Ecolabelling Min Yu, Assistant Professor, University of Portland, 5000 N. Willamette Blvd., Portland, OR, 97203, United States of America, yu@up.edu, Jonas Floden, Anna Nagurney We develop a competitive supply chain network model for fashion that incorporates ecolabelling. We capture the individual profit-maximizing behavior of the fashion firms which incur ecolabelling costs with information associated with the carbon footprints of their supply chains revealed to the consumers. Consumers, in turn, reflect their preferences for the branded products of the fashion firms through their demand price functions, which include the carbon emission information.

237