INFORMS 2021 Program Book

INFORMS Anaheim 2021

MD31

2 - Loot Box Pricing and Design Xiao Lei, Columbia University, New York, NY, 10027-6601, United States, Ningyuan Chen, Adam Elmachtoub, Michael L. Hamilton Online games garner annual revenues in the billions, more than half of which is from purchases of virtual items to be used by the player in the game. One popular way to sell in-game items are via loot boxes, which are random bundles of virtual items, the contents of which are revealed after purchase. We consider how to design loot boxes selling strategies, and compare them with bundle selling and separate selling. We show that in an asymptotic regime, carefully designed loot box strategies can garner as much revenue as bundle selling while inheriting many nice properties of separate selling. Our result and discussion give insights to customers, sellers and regulators. MD33 CC Room 209A In Person: Resilient Infrastructure and Community Networks General Session Chair: Alexander Gilgur, Stevens Institute of Technology, Sunnyvale, CA, 94087-2939, United States 1 - Efficient Solution Approaches to the Isolated Community Evacuation Problem Klaas Fiete Krutein, University of Washington, Seattle, WA, 98117, United States, Anne Goodchild The Isolated Community Evacuation Problem (ICEP) is a recently introduced route optimization model that aims at minimizing the evacuation time for communities that have no road-based evacuation routes and rely on alternative transportation resources for evacuation. The stochastic version of the problem allows for making planning decisions for such events considering demand uncertainty through a set of disaster scenarios. However, since this adds additional complexity and computational effort, we present different approximate solution approaches to the ICEP that reduce the solution time and allow for better use in practice for evacuation planning and show the effect on example problems of varying sizes. 2 - Equitable and Sustainable Energy Transitions Destenie S. Nock, Carnegie Mellon University, Pittsburgh, PA, 15207-1120, United States In the fight against climate change countries have set strong electricity sector decarbonization targets. However, there is uncertainty regarding whether these policies will exacerbate social inequities, and how they will impact environmental sustainability across different income groups. Currently, most electricity planning models determine the least cost option, without considering how the recommended pathways impact distributional equity. This research will explore the sustainability and equity trade-offs between different energy transition pathways for the US. Specifically we tie a national least cost optimization model with and equity analysis. We show how decarbonization targets impact energy equity objectives. 3 - Social Cohesion and Emotion Analysis of News and Tweets During 2020 Wildfires: A Case Study Alexander Gilgur, Stevens Institute of Technology, Sunnyvale, CA, 94087-2939, United States, Jose E. Ramirez-Marquez Wildfires are a fact of life in California, from San Diego to Mount Shasta. We used social and public media to analyze emotions, social cohesion, and resilience in the cities of the San Francisco Bay Area, CA before, during, and immediately after California wildfires of 2020. The effects of interactions with COVID and protests of 2020 have been analyzed as well. MD34 CC Room 209B In Person: Service Science Best Cluster Paper Competition (III) Award Session Chair: Pnina Feldman, Boston University, Boston, MA, 02215, United States 1 - Tax-Induced Inequalities in the Sharing Economy Yao Cui, Cornell University, Ithaca, NY, 14853-6201, United States We use a machine learning (causal forest) method to empirically study the heterogeneous treatment effects of the occupancy tax policy on Airbnb. We find that the tax adversely affects residential listings more than commercial listings, suggesting that the current tax policy may over-penalize the wrong type of listings. We further show that this unintended consequence is caused by customers’ discriminatory tax aversion. We then conduct prescriptive analytics regarding how hosts should optimally adjust prices in response to the tax and

MD31 CC Room 208A In Person: Data-driven Methods for Systems Engineering General Session

Chair: Luis Javier Segura, Buffalo, NY, 14228, United States 1 - An Adaptive Data-driven Kernel for Blind Image Deblurring Sajjad Amrollahi Biyouki, The University of Tennessee, Knoxville, TN, United States, Hoon Hwangbo Blind Image deblurring tries to estimate blurriness and recover a latent image out of a blurred image. This process, as being an ill-posed problem, requires imposing restrictions either on the latent image or a blur kernel representing blurriness. Different from recent studies that impose some priors on the latent image, this research explicitly formulates the structure of the underlying kernel where the structure itself is adaptive to data, which enables modeling nearly non-parametric shape of blurriness. When applied to the recovery of satellite images, the recovered images show the superiority of the proposed method to other state-of- the-art approaches. 2 - Surface Temperature Monitoring in Liver Procurement via Time- vertex Signal Processing Sahand Hajifar, University at Buffalo, Buffalo, NY, United States, Hongyue Sun Accurate evaluation of liver viability during its procurement is a challenging issue. Recently, people have started to investigate the non-invasive evaluation of liver viability during its procurement using the liver surface thermal images. However, existing works attempt to evaluate quality of the liver by extracting either temporal temperature variation or spatial temperature variation. The objective of this study is to jointly extract spatiotemporal (belonging to both space and time) variations to evaluate quality of the liver. To achieve this objective, we use techniques from time-vertex signal processing. In particular, we use joint Fourier transform (JFT) to extract features that contain information from both time and space domains. Then, we use a high-dimensional control chart to monitor the features and estimate the change point. 3 - Inkjet Printing Droplet Evolution Prediction via Tensor Time Series Luis Javier Segura, University at Buffalo, Buffalo, NY, United States, Zebin Li, Luis Javier Segura, Hongyur Sun Droplet behaviors substantially determine the quality of the produced products in the Inkjet Printing (IJP). The droplet formation mechanism (i.e., droplet evolution) understanding is fundamental for the process performance. This work investigates droplet evolution prediction via Tensor Time Series analysis. The method learns the spatial-temporal relationships by joining the force of Tensor Graph Convolutional Network (TGCN) and Tensor Recurrent Neural Network (TRNN). The method is tested in experimental and simulated droplet evolution data in the IJP process. 4 - Challenging Research Problems in the Automotive Industry Arman Sabbaghi, Purdue University, West Lafayette, IN, 47907- 2067, United States Novel challenges in the automotive industry have led to fundamentally new research opportunities in quality, statistics, and reliability. The panelists in this session will discuss the new research problems that they are investigating in their work in the automotive industry. MD32 CC Room 208B In Person: New Challenges in Pricing and Revenue Management General Session Chair: Adam Elmachtoub, Columbia University, New York, NY, 10027- 3241, United States Co-Chair: Xiao Lei, 1 - Menu Design of a Bipartite Matching Queueing System With Strategic Users Lisa Hillas, University of Chicago, Chicago, IL, United States, Rene A. Caldentey, Varun Gupta In this talk, we explore the optimal design of matching topologies for a multi-class multi-server queueing system under a FCFS-ALIS service discipline. We investigate the performance of the system from the perspective of a central planner who must design a menu of service classes, which are defined by the subset of servers that can serve each class. Customers are heterogeneous on their preferences over servers and self-select the service class to join.

82