Informs Annual Meeting 2017

SA32

INFORMS Houston – 2017

SA31

SA29

350E Methods and Models for Big Network Data Sponsored: Artificial Intelligence Sponsored Session Chair: Bin Zhang, PhD, University of Arizona, Tucson, AZ, 85721, United States, binzhang@arizona.edu 1 - When will a Repost Cascade Settle Down? Chi Chen, Tsinghua University, Beijing, China, chenchi14@mails.tsinghua.edu.cn Repost cascades play a critical role in information diffusion on social media. Substantial previous work has studied various aspects of repost cascades such as growth, burst and recur, however, when a repost cascade will settle down remains to be an open problem. Existing models cannot be directly applied to solve the problem because of sensitiveness of feature based models and assumption of point process based models. In this paper, we propose a novel definition “settling time” and develop a model to get rid of the restriction in previous studies. We conduct experiments on Sina Weibo dataset whose results show that our model achieves over 10% performance gain than the previous approaches. 2 - Estimating External Motivating Factors in Interorganizational Social Media: Peer Effects and Organizational Influences Bin Zhang, Assistant Professor, University of Arizona, 1130 E. Helen St, Rm 430Z, Tucson, AZ, 85721, United States, binzhang@arizona.edu, Kexin Zhao, Xue Bai Interorganizational social media (IOSM) enable professionals belonging to different organizations to share knowledge online. Knowledge sharing requires social interactions within and beyond the community boundary. Thus influences from external entities likely play an important role in encouraging individuals’ participation. In this research, we study external motivating factors generated from peer effects and organizations. We apply a novel econometric identification method to analyze a unique dataset collected from an IOSM. We find that external motivating factors from peers and organizations are influential in determining community participation. 3 - Are Electronic Cigarettes a Safer Substitute for Cigarettes for Asthma Patients a Network Based Research Wenli Zhang, University of Arizona, Tucson, AZ, United States, wenlizhang@email.arizona.edu, Sudha Ram Asthma affects millions of US adults. Among all the asthma triggers, one of the biggest risk factors is smoking. However, smoking is one of the most difficult addictions to break. It is believed that the use of Electronic Nicotine Delivery Systems (ENDS) may be considered as a low risk substitute for tobacco use for asthma patients. Studies show that people exposed to an ENDS post on social media may be more easily persuaded if the message is from credible sources (e.g., friends or reliable news media), which may have influence on the adoption of ENDS. The research objective of this study is to explore the key conversations and the propagation of ENDS usage among asthma users on social media. 4 - A Model-free Solver for Arbitrary Graph Problems: Predicting Solutions with Deep Learning Stefan Feuerriegel, University of Freiburg, Graph problems require extensive manual effort due to the mathematical derivation of exact solvers or heuristics. As an alternative, we propose a deep learning methodology that can approximate solutions without the need for a mathematical model of the problem. Our innovative approach merely requires problem-solution pairs, based on which it then learns the underlying mechanism in order to solve the problem. The key is here a memory network for making predictions from arbitrary data structures. This work studies its computational performance and derives bounds. Platz der alten Synagoge, Freiburg, 79098, Germany, stefan.feuerriegel@is.uni-freiburg.de, Nil-Jana Akpinar

351A Nicholson Student Paper Competition I Invited: Nicholson Student Paper Prize Invited Session

Chair: Hayriye Ayhan, Georgia Institute of Technology, School of Industrial & Systems Eng, 755 Ferst Dr NW, Atlanta, GA, 30332-0205, United States, hayhan@isye.gatech.edu Co-Chair: Cole Smith, Clemson University, 110 Freeman Hall, Clemson, SC, 29634, United States, jcsmith@clemson.edu 1 - Dynamic Matching for Real-time Ridesharing Erhun Ozkan, University of Southern California, Los Angeles, CA, United States, erhunozkan@gmail.com In a ridesharing system such as Uber or Lyft, arriving customers must be matched with available drivers. We propose to base the matching decisions on the solution to a continuous linear program (CLP) that accounts for (i) the differing arrival rates of customers and drivers in areas of the city, (ii) how long customers are willing to wait for driver pick-up, and (iii) the time-varying nature of all the aforementioned parameters. 2 - Tractable Distributionally Robust Optimization with Data Zhi Chen, National University of Singapore, Singapore, chenzhi.james@gmail.com We present a framework for distributionally robust optimization that could encompass a variety of statistical information including constraints on expecta- tion, conditional expectation, and disjoint confidence sets with uncertain proba- bilities defined by phi-divergence. We also show that the Wasserstein-based ambiguity set has an equivalent formulation via our ambiguity set, which enables us to tractably approximate a Wasserstein-based distributionally robust optimization problem with recourse. 3 - On Deterministic Reformulations of Distributionally Robust Joint Chance Constrained Optimization Weijun Xie, Georgia Institute of Technology, Atlanta, GA, wxie33@gatech.edu A joint chance constrained optimization problem (CCP) involves uncertain con- straints that are jointly required to be satisfied with probability exceeding a pre- specified threshold. In a distributionally robust joint CCP (DRCCP), the joint chance constraint is required to hold for all probability distributions of the sto- chastic parameters from a given ambiguity set. We consider DRCCP involving convex nonlinear uncertain constraints and an ambiguity set specified by convex moment constraints.

4 - Multi-agent Online Learning under Imperfect Information:Algorithms, Theory and Applications Zhengyuan Zhou, Stanford University, Stanford, CA, zyzhou@stanford.edu

We consider multi-agent online learning under imperfect information, where the reward structures of agents are given by a general continuous game. After intro- ducing a general equilibrium stability notion for continuous games, we examine the online mirror descent (OMD) algorithm and show that its last iterate con- verges to variationally stable Nash equilibria under certain assumptions. We also propose a delayed mirror descent algorithm that relies on the repeated leveraging of past information.

SA32

351B Drone Delivery Systems – I Invited: Invited Drone Delivery Systems Invited Session Chair: James F. Campbell, University of Missouri-St Louis,

Saint Louis, MO, 63121-4499, United States, campbell@umsl.edu 1 - A Branch and Bound Approach to the TSP with Drone and Related Heuristics Stefan Poikonen, PhD Candidate, University of Maryland-College

Park, 3406 Tulane Dr., Apt. 22, Hyattsville, MD, 20783, United States, spoikone@math.umd.edu, Bruce L.Golden

The traveling salesman problem with a drone (TSP-D) is a hybrid truck and drone model of delivery. A drone may ride atop a truck. Trucks may deliver packages directly. Alternatively, the drone may carry a package from the truck to a customer then return to the top of the truck to swap batteries. This system benefits from the drone’s use of the crow-fly metric and parallelization of work, yet mitigates concerns about limited drone range. Computational results, both exact and heuristic, will be shown. We greatly expand the size of problems that may be solved optimally. Comparisons to other solution methods in the literature will be made.

13