Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

WD72

3 - BitExTract: Interactive Visualization for Extracting Bitcoin Exchange Intelligence Siyuan Liu, Pennsylvania State University, University Park, PA, United States The emerging prosperity of cryptocurrencies, for example, Bitcoin, has come into the spotlight during the past few years. In this paper, we present BitExTract, an interactive visual analytics system, which, to the best of our knowledge, is the first attempt to explore the evolutionary transaction patterns of Bitcoin exchanges from two perspectives, namely, exchange versus exchange and exchange versus client. The effectiveness and usability of BitExTract are demonstrated through three case studies with novel insights and further interviews with domain experts and senior Bitcoin practitioners. n WD70 West Bldg 106B Practice - Computer Science – Applications to OR Contributed Session Chair: Hiroko Okajima, Towson University, 8000 York Road, Towson, MD, 21252-0001, United States 1 - On the Consistency and Computation of Maximum Likelihood Estimators for Multivariate Hawkes Processes Anran Hu, PhD Student, UC Berkeley, Albany, CA, 94706, United States, Xin Guo, Renyuan Xu, Junzi Zhang We establish the first consistency result for the maximum likelihood estimators (MLEs) of multivariate Hawkes processes (MHPs) with general linear intensity processes, under mild and verifiable conditions. We also develop an alternating minimization type algorithm with guaranteed global convergence to the set of critical/stationary points. The performance of the proposed algorithm on both synthetic and real-world data is reported, showing the advantage of our approach. 2 - Balance Optimization Subset Selection with Multiple Treatment Levels Hee Youn Kwon, Northwestern University, 2211 Campus Drive, Evanston, IL, 60208, United States, Jason J. Sauppe, Sheldon H. Jacobson Matching and Balance Optimization Subset Selection (BOSS) are two frameworks for causal inference with observational data. These methods find an estimate for the treatment effect either by matching each treatment unit to a control unit or by finding a control group that is balanced with respect to the treatment group. However, when there are multiple - more than two - treatment levels, methods of comparing two levels are not directly applicable. The goal of this talk is to generalize the BOSS framework which was developed under a binary treatment setting to a multi-treatment setting. The BOSS estimator is computationally compared to the matching estimators. 3 - A Balancing Block Based Formulation for the Police Districting Problem: The Case of Antofagasta Police Department Evelyn Arrey, Universidad Catolica del Norte, Avenida Angamos 0610, Building Y1, Antofagasta, Chile, Hernan Caceres Preventive security is one of the most critical social priorities of a community, for which a typical strategy is to design patrol areas. Such designs consist on dividing a city according to one or several features of its neighborhoods, that are area- based indicators in most cases. In this work, we propose a mixed integer program that splits an urban area by balancing a cumulative block-based indicator among a given number of sub-divisions. For our numerical example, we present a case study of the Police Department of Antofagasta, Chile. We compare our results with the division that the city currently uses, that is based on the nation-wide program Quadrant Plan. 4 - Can Group Giving Boost Contribution?

n WD71 West Bldg 106C Combinatorial Optimization over Graphs Sponsored: Computing Sponsored Session Chair: Illya V. Hicks, Rice University, Houston, TX, 77005-1892, United States 1 - Catching a Robber Quickly with Few Cops Boris Brimkov, Rice University Cops and Robbers is a pursuit-evasion game played on a graph, between a team of cops and a single robber. The cops and the robber take turns moving to adjacent vertices; the goal for the cops is to capture the robber by occupying the same vertex as the robber, and the goal of the robber is to avoid capture. We consider the cop-throttling number of a graph G, which is defined as the minimum possible value of (k + the minimum number of turns needed for k cops to capture the robber on G over all possible games). This talk will provide some tools for computing and bounding the cop-throttling number, and relate it to other graph parameters. 2 - SAT and SMT Encodings of the Zero Forcing Problem Derek Mikesell, Rice University, Houston, TN, United States Zero forcing is an iterative graph coloring process in which a colored vertex forces a neighboring vertex to be colored if it is the only uncolored vertex in its neighborhood. The Zero Forcing Problem asks for the smallest set of initially colored vertices such that the entire graph becomes colored under this rule. In this work, computational methods are proposed for the Zero Forcing Problem and related variants. The methods are based on a Boolean Satisfiability and Satisfiability Modulo Theories encoding of an infection model. The methods yield competitive results to state-of-the-art integer programming and combinatorial algorithms. 3 - A Set Cover Formulation for Power Domination in Graphs Logan Smith, Rice University, Houston, TX, United States To observe electrical networks, sensors are installed at power substations. Sensors directly observe adjacent substations and may indirectly observe others. In graphs representing power grids, vertices corresponding to substations with sensors are initially colored; the power domination color rule indicates which substations are observable. Previous computational methods rely on modeling the coloring rules and vertex coloring sequences. Here, an alternative exploiting graph structures that prevent colorings is introduced. Unlike previous work, this approach permits a separation algorithm. Computational experiments show significant runtime improvement for the proposed algorithm. n WD72 West Bldg 211A Practice- Operations/ Marketing Interface Contributed Session Chair: Jinpeng Xu, Xidian University, School of Economics and Management, Xi’an, 710126, China 1 - Supply Network Effects on the Scalability of Online Grocery Retail Platforms Lina Wang, Student, Arizona State University, Tempe, AZ, 85257, United States, Timothy Richards, Elliot Rabinovich A grocery retail platform acts as a two-sided market connecting consumers and suppliers. Its scalability depends on the type and the number of vendors participating in the platform. We study empirically indirect network effects on scalability in one of these platforms. Whether these effects are positive will depend on the impact that the number and types of vendors participating in the platform have on consumers’ demand. Results obtained from a consumer demand model and a supply provision equilibrium model confirm the existence of a positive indirect network effect on the scalability of these platforms. 2 - Keep the Buzz and Binge on Optimal Content Release Strategies Clint Ho, Imperial College Business School, London, SW2 7AZ, United Kingdom, Esma Koca, Wolfram Wiesemann Video-on-demand platforms unveiled a game-changing media consumption phenomenon, binge-watching, with the release of all episodes of series at once (entire season at once policy) and hence generate a binge-watching driven buzz among viewers. As opposed to this, traditional strategy (episodic release strategy) creates a viewer-generated chatter around the story, commonly called as the water-cooler effect. We argue that a hybrid and a personalized strategy may activate both levers and investigate three aspects of the optimal content release: (1) optimal release of a series, (2) optimal order of multiple series, and (3) discrimination on the content across heterogeneous viewers.

Hiroko Okajima, Assistant Professor, Towson University, 8000 York Road, Towson, MD, 21252-0001, United States

Group giving (collective giving) has been gaining popularity in practice, but little has been studied in literature. This study supplements the existing literature by providing empirical results on group giving. Our laboratory experiment reveals that group giving is especially effective in increasing giving rates when it is used in combination with stepwise rebates.

517