Informs Annual Meeting 2017

TC50

INFORMS Houston – 2017

TC50

shares are disproportionately more likely to be recommended. We propose a novel method that recommends items with probabilities proportional to each item’s sales or views among similar users. Using a dynamic simulation analysis, we show that our approach dramatically increases both aggregate and individual- level recommendation diversity while simultaneously increasing recommendation accuracy. 2 - Comparing the Effects of Disparate Sources of Information on Consumers’ Product Selection: A Process-oriented View Jingjing Zhang, Indiana University, 1309 E 10th St, HH4143, Bloomington, IN, 47405, United States, jjzhang@indiana.edu, Binny Samuel, Vijay Khatri The objective of this project is to develop an understanding of consumers’ information integration process in the online shopping setting. We are interested in identifying the sequence in which consumers use to integrate different sources of information (i.e., product description, online word-of-mouth, and personalized recommendations) in their product selection process. 3 - Recommender System and Cross-selling: Monopoly and Duopoly Abhijeet Ghoshal, University of Louisville, Room 377, College of Business, Louisville, KY, 40292, United States, abhijeet.ghoshal@louisville.edu, Sumit Sarkar, Vijay Mookerjee Recommendations provided by firms change how customers search products by helping customers find their preferred products. In the process, firms learn the preferences of customers and make cross-selling offers. Considering these factors, we study how firms price products in monopolistic and duopolistic markets and highlight the role of cross-selling. In duopoly, cross-selling may compel the firm with inferior recommender system to reduce its price when the superior firm improves its system, a counter-intuitive result since improvement of the system of the latter firm decreases competition. 361E Energy Contributed Session Chair: Deepak Rajan, Lawrence Livermore National Laboratory, Livermore, CA, United States, rajan3@llnl.gov 1 - Impact of Various Cascading Assumptions on Defensive Investments Sinan Tas, Assistant Professor, University of Wisconsin-Platteville, 1 University Plaza, Platteville, WI, 53818, United States, sinantas@gmail.com Cascading failure is often overlooked when analyzing how vulnerable power grids, or similar capacity-constrained networks, are. In this study, we first model cascading failure probabilistically, and then analyze how various cascading assumptions about the network and the attacker can change our investments to make the system more secure or resilient. 2 - Optimal Allocation of HVDC Interconnections for the Exchange of Energy and Reserve Capacity Services TC52 The European market structure separates energy and reserve trading leading to inefficient utilization of HVDC interconnections, since their capacity must be ex- ante allocated between these services. To improve market efficiency, we propose a decision-support tool that enables an implicit temporal coupling of the different trading floors, using as control parameters the inter-area transmission allocation between energy and reserves and the reserve requirements. The proposed mechanism, formulated as a stochastic bilevel program, is aligned with the existing market and reduces expected system cost for high shares of intermittent renewables. 3 - Efficient Design for Short Term Markets in the European Electricity System - First Results from a Stochastic Multi Stage Energy System Model Frieder Borggrefe, German Aerospace Center, Pfaffenwaldring 38-40, Stuttgart, 70563, Germany, frieder.borggrefe@dlr.de This paper introduces an integrated model for short-term markets within the European electricity system. Aim of the model is to analyze and compare future market designs. The paper shows first results from a simplified commitment and dispatch model including day-ahead, intraday & balancing markets and imbalance settlement rules.The second part compares market designs and influence of uncertainty on market decisions. Due to complexity of the model it is difficult to scale it to the full European market. The paper outlines next steps to expand this model and discusses challenges when applying it HPC. The model results are part of the BEAM-ME project. Stefanos Delikaraoglou, Technical University of Denmark, Akademivej, Building 358, Kgs. Lyngby, 2800, Denmark, sdelikaraoglou@gmail.com

361C Connected and Automated Transportation Sponsored: TSL, Urban Transportation Sponsored Session

Chair: Husain Abdul Aziz, PhD, Oak Ridge National Laboratory, Oak Ridge National Laboratory, 1 Bethel Valley Road, Oak Ridge, TN, 37831, United States, azizh@ornl.gov 1 - Simulating the Operations of Shared Autonomous Vehicle Fleets Krishna Gurumurthy, University of Texas-Austin, Stop C1700, Austin, TX, 78712, United States, na, Kara Kockelman Simulation of fleets of shared autonomous vehicles (SAVs) exist for various travel settings, using Austin, Texas data. Reliance on all-electric vehicles lowers replacement rates (from 1 SAV for 9 household vehicles to 1-to-5, in some cases) and raises empty-mileage (to 20% of fleet VMT). Dynamic ride-sharing (among strangers) raises replacement rates (to 10:1, for example) and lowers empty-VMT (becoming negative in some cases). Increasing a served region’s size, to enable all trip distances, does not appear to raise empty-VMT, but response times are higher in exurban settings, and untenable with all-electric fleets. 2 - Automated Vehicle Performance at Intersection. Integration of Empirical Evidence with Simulation Ramin Arvin, Graduate, University of Tennessee-Knoxville, 325 John D. Tickle Building, Knoxville, TN, United States, rarvin@vols.utk.edu Automated Vehicles (AV) can potentially improve the current state of transportation by reducing human errors, and improving safety and reliability. In this paper, we develop crash scenarios using early empirical evidence from real- life AV-involved crashes in a simulation framework. The Wiedemann car-following model in SUMO simulates both traffic flow and crashes. Therefore it was used to quantify crash rates under different AV market penetration scenarios. 3 - Roads in Transition: Integrated Modeling of a Manufacturer- traveler-road Infrastructure System in a Mixed Autonomous/Human Driving Environment Bo Zou, Assistant Professor, University of Illinois at Chicago, 2073 Engineering Research Facility, 842 W. Taylor Street, Chicago, IL, 60607, United States, bzou@uic.edu, Mohamadhossein Noruzoliaee, Yang Liu This study develops an integrated modeling framework for autonomous vehicle (AV) manufacturer-traveler-road infrastructure system in a futuristic, mixed traffic environment. The modeling system involves manufacturer decision, multiclass traveler decision on vehicle and routing choice, and variable road capacity, and is formulated as an MPCC. To solve the MPCC to global optimality, multiple MILP-based approximation techniques are developed. The effectiveness of the solution approach is demonstrated with extensive numerical experiments in a Singapore case study. 4 - Machine Learning Based Traffic Signal Control Algorithms for Connected and Automated Transportation Husain Aziz, Dr., Oak Ridge National Laboratory, 1 Bethel Valley Road, Oak Ridge, TN, 37831-6017, United States, azizh@ornl.gov The anticipated deployment of the connected and automated vehicles (CAVs) offer the opportunity to leverage connectivity and automation for efficient operations of signalized intersection in transportation networks. CAVs can act as mobile sensors with a unique opportunity for data collection that can be used for real-time optimization of traffic signal settings. Machine learning based signal control techniques are expected to adapt to the changing traffic condition both temporally and spatially. We developed a machine-learning based signal control scheme that leverages the available data in a CAV environment and accounts for energy and mobility. 361D Recommender Systems Sponsored: Information Systems Sponsored Session Chair: Konstantin Bauman, New York University, New York, NY, 10012, United States, kbauman@stern.nyu.edu 1 - Boosting Recommendation Diversity through Stochastic Item Selection Kartik Hosanagar, University of Pennsylvania, 3730 Walnut Street, TC51

500 Jon M.Huntsman Hall, Philadelphia, PA, 19104-6340, United States, kartikh@wharton.upenn.edu, Alex Miller

The standard k-nearest neighbor collaborative Filtering algorithm is well known to exhibit systematic popularity bias, by which products with higher market

354