Informs Annual Meeting 2017

MC49

INFORMS Houston – 2017

MC49

3 - A Dual Machine Learning-model Based Framework for Privacy Analysis of Traffic Christian Claudel, University of Texas-Austin, TX, 78750, United States, christian.claudel@utexas.edu In this talk, we consider the problem of tracking vehicles generating anonymous GPS tracks. Using an Hamilton Jacobi formulation of traffic flow, we pose the problem of reidentifying anonymous tracks as a Mixed Integer Linear program (MILP). This MILP results is used as an input to an Artificial Neural Network (ANN) classifier that reidentifies vehicles from anonymous tracks. This framework is validated using experimental location data, and shows that 94% reidentification is possible, considerably above random reidentification (50%) or naive reidentification based on the kinematics of vehicles (88%). 4 - An Algorithm for Preserving the Privacy of Sharing Transport Service Route Data Yueshuai He, New York University, New York, NY, United States, yh1995@nyu.edu, Joseph Y j Chow Despite the increasing relevance of private transport operators as Mobility-as-a- Service in the success of smart cities, desire for privacy in data sharing limits collaborations with public agencies. We propose an original model and algorithm that circumvents this limitation, by designing a diffusion of the data—in this case, service tour data—such that passenger travel times remain reliable to the recipient agency. The Tour Sharing Privacy Design Problem is formulated as a nonlinear programming problem that maximizes entropy. An algorithm based on formulation of an equivalent kth best traveling salesman problem is evaluated using several network instances. 361D Field Experiment in Online Marketplace and Social Network Sponsored: Information Systems Sponsored Session Chair: Tianshu Sun, University of Southern California, Los Angeles, CA, 90089, United States, tianshus@marshall.usc.edu 1 - Learning from Inventory Information: Field Evidence from Amazon Ruomeng Cui, Goizueta Business School, Emory University, 1300 Clifton Rd, Atlanta, GA, 30322, United States, ruomeng.cui@gmail.com, Dennis Zhang, Achal Bassamboo Many online retailers provide real-time inventory availability information. Customers can learn from the inventory level and update their beliefs about product quality. Based on a unique setting from Amazon lightning deals, which displays the percentage of inventory consumed in real time, we explore whether and how consumers learn from inventory availability information. We run randomized field experiments on Amazon and run a panel data analysis. We find evidence of consumer learning from inventory information: a decrease in product availability causally attracts more sales in the future; in particular, a 10% increase in past sales leads to a 2.08% increase in cart add-ins in the next hour. 2 - Dynamic Personalized Targeting with Hidden Use Engagement Stages: Mobile Tapstream Data and Field Experiment Yingjie Zhang, Carnegie Mellon University, Pittsburgh, PA, United States, yingjie2@andrew.cmu.edu, Beibei Li, Xueming Luo, Xiaoyi Wang Low engagement rate and high attrition rate have been major challenges for the success of mobile apps. To date, little is known towards how companies can improve user engagement and business revenues through designing effective in- app pricing strategies. We propose a structural model by accounting for time-varying nature of engagement and consumer forward-looking behavior. We analyze mobile tapstream data from a popular mobile reading app. Our results enable us to tailor optimal pricing strategy to each consumer based on their engagement status. Interestingly, we found such engagement-specific pricing strategy leads to lower average price for consumers and higher overall business revenues. 3 - Social Influence Across Products: A Large-Scale Randomized Experiment in Social Advertising Shan Huang, Massachusetts Institute of Technology, 100 Main Street, # 427, Cambridge, MA, 02142, United States, shanh@mit.edu, Sinan Aral, Jeffrey (Yu) Hu, Erik Brynjolfsson Spending on social advertising is increasingly dramatically. Unfortunately, large- scale experimental evidence of the lift from social ads almost always focuses on a single product at a time. As a result, we know little about how social influence and the effectiveness of social advertising vary across product categories. We, therefore conducted a randomized field experiment measuring social influence and social ad effectiveness across 71 products in 25 product categories among a random sample of more than 37 million users of WeChat Moments Ads. MC51

361B Influence of Human Behaviour on Project Success: Theories, Models and Practices Invited: InvitedBehavioral Aspects of OR Invited Session Chair: Sijun Bai, Management School, Northwestern Polytechnical University, Xi’an, China, baisj@nwpu.edu.cn Co-Chair: Lin Wang, Management School, Northwestern Polytechnical University, Xi’an, China, linwang.nwpu@gmail.com 1 - Research on Emergency Decision Management of Chemical Accidents Based on Game Theory He Zhai, Northwestern Polytechnical University, Xi’an, China, zhaihe0218@163.com, Mengna Guo, Yuntao Guo With the arrival of China’s economic and social transition period,the complexity and dynamics of various projects are gradually increasing, which makes the project management more difficult. In recent years, the safety accidents and unexpected events in Shaanxi province emerge in endlessly,and the “conspiracy of optimism” is the main factor leading to the failure of decision-making. Based on the game theory, this paper analyzes the reasons of the “conspiracy of optimism” phenomenon in the process of emergency decision making and a model will be constructed based on this problem, which may provide a reference for the improvement of the emergency decision-making management. 2 - Project Portfolio Implementation under Uncertainty: A Behavioral Perspective Lin Wang, Northwestern Polytechnical University, Xi’an, China, linwang.nwpu@gmail.com Lin Wang, Institutes of science and development, Chinese Academy of Sciences, Beijing, China, linwang.nwpu@gmail.com, Sijun Bai As a social system highly influenced by decision makers’ behaviors, the management of project portfolios should highlight behavioral factors instead of adhering to rational and normative plans. This paper investigates the uncertainties generated by behavioral biases (e.g. planning fallacy, strategic misrepresentation and optimism bias), how should decision makers respond to these uncertainties, and the impact of individual and interactive biases (e.g. lies and delays) on the monitoring and control process. System dynamics is applied to model the complex system and provide the foresight of each decision, and an example is illustrated to validate the model. 361C Privacy in Urban Transportation Sponsored: TSL, Urban Transportation Sponsored Session Chair: Christian Claudel, University of Texas-Austin, Austin, TX, United States, christian.claudel@utexas.edu 1 - Privacy Aware Traffic Monitoring in Work Zones Daniel Work, University of Illinois, Urbana, IL, United States, dbwork@illinois.edu, Yanning Li Despite the increasing popularity of Lagrangian sensors, fixed location traffic sensors are still irreplaceable in certain safety-critical applications, such as monitoring traffic in work zones. However, the high cost and privacy concerns of conventional fixed video-based monitoring systems create economic and policy challenges for dense sensor deployments. This talk advocates the use of a low- resolution passive infrared camera to reduce both energy demands and privacy challenges. To extract vehicle counts and speeds from the sensor, an unsupervised machine learning algorithm is proposed. Results from recent field experiments will be discussed. 2 - Differentially Private Road Traffic Estimation Jerome Le Ny, Polytechnique Montreal, Montreal, QC, H3T1J4, Canada, jerome.le-ny@polymtl.ca Urban transportation systems rely on an increasing number and variety of sensors capturing our activities in order to optimize their operations, which raises important privacy concerns. Naive data sanitization techniques such as removing names from location traces are known to provide insufficient protection. This talk gives an introduction to differential privacy, a state-of-the-art formal tool for privacy preserving data analysis, and illustrates its application to real-time traffic estimation with static sensors and floating car data. MC50

220