2016 INFORMS Annual Meeting Program

MC39

INFORMS Nashville – 2016

MC37 205C-MCC

2 - Positive Impact Of Graphical Visualization Of Discussion Forums On Collaborative Learning Jacqueline Ng, Northwestern University, 2145 Sheridan Road, C-230, Evanston, IL, 60208, United States, jacqueline.ng@u.northwestern.edu, Seyed Iravani Widespread internet connectivity has increased the popularity of online delivery of course content. With the rise of online courses, e.g., MOOCs, there is an increasing need to create opportunities for learners to interact and exchange ideas. Dynamic online discussion forums can accomplish these goals. We use visualization techniques to design a novel graphical interface for discussion forums that presents posts as nodes and replies as edges connecting nodes. By comparing the effectiveness of graphical and text-based discussion forums, we find that the graphical interface promotes higher levels of both activity and interactivity, creating increased engagement in online discussions. 3 - Public Reactions To Supply Chain Events: A Twitter Sentimental Analysis Event Study David Wuttke, EBS University, Burgstr. 5, Oestrich-Winkel, 65375, Germany, david.wuttke@ebs.edu, Christoph Schmidt, H. Sebastian Heese We conduct sentiment analysis on Twitter data to evaluate public reactions to supply chain events. 4 - Effectiveness Of Network-Based Evacuation Warning Dissemination: An Experimental Investigation Sulian Wang, Tsinghua University, 30 Shuangqing Road, Haidian, Beijing, 100084, China, wangsulian13@mails.tsinghua.edu.cn, Chen Wang Effective risk communication with the general public plays a vital role in emergency preparedness and response. Spontaneous dissemination of warning messages in the decentralized channel (e.g., through online social network) is shown to be an efficient way of complementing the traditional channels such as television and radio. We model the individual willingness to spread warning messages as a function of their past experiences and trust of the information source, which is determined by both the false positive and false negative rates of historical warnings. We validate our model by lab experiments and simulation. MC39 207A-MCC Applied Probability and Optimization II Sponsored: Applied Probability Sponsored Session Chair: Jiaming Xu, The Wharton School of the University of Pennsylvania, jiamingx@wharton.upenn.edu 1 - Low-rank Estimation: Why Non-convex Gradient Descent Works Yudong Chen, Cornell University, Ithaca, NY, United States, yudong.chen@cornell.edu Many problems in statistics involve fitting a low-rank matrix to noisy data. A popular approach to the resulting rank-constrained optimization is SDP relaxation, which does not scale to large problems. We instead consider gradient descent over the low-rank space. This approach is scalable, but convergence was unclear due to non-convexity. We develop a unified framework characterizing its convergence and statistical properties. Our results provide insights to why we expect non-convex methods to work in general, and yield global guarantees for linear convergence in various concrete problems. Our framework handles arbitrary ranks, noise and constraints, and does not require sample splitting. 2 - Scaled Least Squares Estimator For Glms Murat A. Erdogdu, Stanford University, erdogdu@stanford.edu We study the problem of efficiently estimating the coefficients of generalized linear models (GLMs) in the large-scale setting where the number of observations n is much larger than the number of predictors p, i.e. n>>p>>1. We show that in GLMs with random design, the GLM coefficients are approximately proportional to the corresponding ordinary least squares (OLS) coefficients.Using this relation, we design an algorithm that achieves the same accuracy as the maximum likelihood estimator through iterations that attain up to a cubic convergence rate, and that are cheaper than any batch optimization algorithm by at least a factor of O(p). 3 - Reinforcement With Fading Memories Se-Young Yun, Los Alamos National Lab, Los Alamos, NM, United States, yunseyoung@gmail.com, Kuang Xu Can one make good decisions despite having a faulty memory? We study a continuous-time action-rewards process, where an agent is to select a sequence of actions from a finite set of alternative, and during the period when action $k$ is selected, she accrues discrete rewards according to a Poisson process of rate $\lambda_k$. However, each unit of reward randomly “fades” from the agent’s memory at rate $n^{-1}$. We analyse a simple reward matching rule: the new action is sampled from a distribution proportional to the recallable rewards associated with actions chosen in the past.

Socially Responsible Business Models Sponsored: Manufacturing & Service Oper Mgmt, Sustainable Operations Sponsored Session Chair: Serguei Netessine, INSEAD, Singapore, Singapore, serguei.netessine@insead.edu 1 - To Sell And To Provide? The Economic And Environmental Implications Of The Auto Manufacturer’s Involvement In The Car Sharing Business Ioannis Bellos, George Mason University, ibellos@gmu.edu Mark Ferguson, Beril L Toktay Motivated by the involvement of Daimler and BMW in the car sharing business we consider an OEM who contemplates introducing a car sharing program. The OEM designs its product line by accounting for the trade-off between driving performance and fuel efficiency. We determine the efficiency of the vehicles offered and we characterize the effect on the OEM’s Corporate Average Fuel Economy (CAFE) along with the economic and environmental implications. 2 - Optimal Allocation Rules With Waste Considerations Sara Rezaee Vessal, HEC Paris, Jouy en josas, France, sara.rezaee-vessal@hec.edu, Sam Aflaki, Dimitrios Andritsos We study capacity allocation of a scarce and perishable product among stockout- averse retailers that face stochastic demand. We focus on two commonly practiced allocation mechanisms and using a dynamic model characterize the conditions under which each allocation mechanism performs superior from a waste and profit point of view. 3 - Child Labor In Supply Chains: An Empirical Investigation Sameer Hasija, INSEAD, sameer.hasija@insead.edu, Hsiao-Hui Lee, Niyazi Taneri Due to increasing globalization, labor malpractices at upstream positions in supply chains directly or indirectly impact many organizations. Moreover, from a social/moral perspective poor labor conditions may have long term adverse effects on society. Lack of visibility in long supply chains hinders our capability in overcoming such issues. In this paper, we generate empirical insights on the drivers of labor malpractices, and child labor in particular. Serguei Netessine, INSEAD, 1 Ayer Rajah Avenue, Singapore, 138676, Singapore, serguei.netessine@insead.edu Jasjit Singh, Nina Teng Companies commonly use philanthropic campaigns to attract and retain customers in the form of charity-linked promotions, where a company donates money to a cause when a customer makes a purchase. Customer-related effects of such promotions remain under-studied, an issue this study investigates using field experiments in an online taxi booking company. Take-up rates for charity-linked promotions were smaller than for discount-based promotions, and also less sensitive to the monetary amount. Although promotion take-up did represent new bookings rather than substitution of non-promotional bookings, there is little evidence of an increase in subsequent purchase frequency. 4 - Philanthropic Campaigns And Customer Behavior: Field Experiments In An Online Taxi Booking Company

MC38 206A-MCC

Social Media Analysis V Invited: Social Media Analytics Invited Session

Chair: Chris Smith, Air Force Institute of Technology, 2950 Hobson Way, 2950 Hobson Way, Wright-Patterson AFB, OH, 45433, United States, cms3am@virginia.edu 1 - App Developers’ Product Offering Strategies In Multi-platform Markets Degan Yu, PhD Candidate, University of Rhode Island, Ballentine Hall, 7 Lippitt Road, Kingston, RI, 02881, United States, yudegan@gmail.com Mobile application (app) developers usually face challenges in deciding product offering choices. In this research we construct an analytical model for product offering problem that app and software developers face in a two-platform environment and the developers offer paid or free app (free app offers advertisement) in each platform. Our findings shed light on some insights into the business practices in industries including mobile apps, computer software, and social media games.

195