INFORMS 2021 Program Book

INFORMS Anaheim 2021

TD03

3 - Shapley-value-based Feature Attribution for Data Privacy Protection Francis Bilson Darku, Research Assistant Professor, University of Notre Dame, Notre Dame, IN, United States, Xinxue Qu, Hong Guo With abundant consumer data, companies can offer better products to customers using AI tools. To protect consumer privacy, several methods have been proposed to reduce identity and disclosure risks. Considering intruders may use other variables in a dataset to infer confidential values, masking or removing the confidential variable(s) only may not be the most effective solution. To address this, we propose a unified framework of feature attribution for data privacy protection. We use a Shapley-value-based approach to fairly allocate the risk and utility of a dataset to the variables contained in that dataset, thereby identifying variables that pose higher risk but contribute less to utility. 4 - Examining the Impact of Yelp’s Elite Squad on Users’ Following Contribution Xunyi Wang, Baylor University, Waco, TX, United States, Sean Sanders, G. Lawrence Sanders As a leading online review platform, Yelp leverages a gamification design, Yelp Elite Squad (YES), to engage its users. A user needs to significantly contribute to the Yelp community to become a YES member. However, it is not clear how the YES members behave after they receive this status. In this study, we use a large volume of Yelp’s data to answer this question. The results based on a dynamic panel analysis show that a YES member will produce more content after obtaining the YES. We further find that if a user receives the YES status for two consecutive years, the most current YES status has a positive effect on users’ future contributions, but this effect is not as strong as the effect of a first-time status recipient. This research contributes to the online user engagement and gamification literature. We also provide practical implications for online community design. 5 - Digital Prophylaxis for Firm Resilience: A Study on Covid-19 Disruption Yuanyang Liu, University of Tennessee, Knoxville, TN, 37934- 4461, United States This article examines whether digitalization increases firm resilience in the presence of a negative exogenous event. Using an extensive online database of job postings, we propose a new measure of firm digitalization that is based on its investment in IT-related human capital. We test the efficacy of firm digitalization as a prophylaxis against the disruption created by the COVID-19 pandemic. Our identification strategy relies on two-way fixed regression and the synthetic control method. We find that higher levels of firm digitalization result in greater firm resilience during COVID-19 disruptions. Our study has implications for managers, shareholders, and regulators as it provides evidence that higher levels of firm digitalization leads to greater firm resilience when faced with negative exogenous shocks. TD04 CC Ballroom D / Virtual Theater 4 Hybrid Network Flow Algorithms in Covariate Balancing Sponsored: Computing Society Sponsored Session Chair: Dorit Simona Hochbaum, University of California-Berkeley, Berkeley, CA, 94720-1777, United States 1 - Complexity and Algorithm of the Integer Minimum Cost Network Flow with Fractional Supplies Problem with an Application to Covariate Balancing Xu Rao, University of California Berkeley (now at Google), Berkeley, CA, 94706, United States, Dorit Simona Hochbaum, Asaf Levin We study the integer minimum cost network flow when some of the supplies are fractional. In the presence of fractional supplies it is impossible to satisfy the flow balance constraints, creating an imbalance. Assigning penalties to the imbalances, we present a polynomial time algorithm to find integer flow that minimizes the sum of the cost of the flow and the imbalance penalty. We also prove that in the presence of a constraint that bounds the imbalance the problem is NP-hard, but efficiently solvable for a fixed number of fractional supplies. This problem is motivated by the application of minimizing proportional imbalance of covariates. 2 - Network Flow Methods for the Minimum Covariate Imbalance Problem Jason Sauppe, University of Wisconsin-La Crosse, La Crosse, WI, United States, Dorit Simona Hochbaum, Xu Rao Given samples of treatment and control units in an observational study, we consider the problem of finding a subset of control units that has minimum imbalance with respect to the treatment sample on two nominal covariates. We present a minimum cost network flow (MCNF) formulation and a maximum flow formulation for solving the problem with the constraint that the control subset size equals the treatment sample size. For other control subset sizes, using a

contribution is rigorous treatment of an important problem that unifies facility location and layout theories with minisum objective and rectilinear distance metric. 3 - A Directional Approach to Gradual Cover Zvi Drezner, California State University Fullerton, Steven G. Mihaylo College of Business and Economic, Fullerton, CA, 92834, United States, Tammy Drezner, Pawel J. Kalczynski The objective of classic cover location models is for facilities to cover demand within a given distance. The gradual (or partial) cover replaces abrupt drop from full cover to no cover by gradual decline in cover. We propose, analyze, and test a new rule for calculating the joint cover of a demand point which is partially covered by several facilities. Every facility covers points within a given radius. Each demand point is defined as a circle with a given radius. The proportion of the demand point covered by a facility is the fraction of the area of the demand circle covered by the facility. The total cover by several facilities is the union of the covered areas. The rule considers the directions of the facilities providing partial cover and not just the proportions of the partial covers. The new approach provided better total cover than the cover obtained by existing procedures. TD03 CC Ballroom C / Virtual Theater 3 Hybrid - Business Analytics in the Age of AI Sponsored: Artificial Intelligence Sponsored Session Chair: Xinxue Qu, University of Notre Dame, Granger, IN, 46530-8209, United States 1 - Animated Live Streamers with AI Assistance on E-commerce Platforms Si Xie, The University of Texas at Dallas, Richardson, TX, United States, Amit Mehra, Zehan Zhao As live streams in e-commerce emerge, some vendors have chosen to deploy AI assisted animated live streamers to broadcast their brands and products. While AI live streamers could help attract consumers without human presence to increase sales, they are costly to develop and are perceived as less persuasive, which might lead to lower sales. In this research, we empirically examine the vendors with AI assisted animated live streamers with their key competitors without it and study the impact on sales and consumer engagement indicators. We also link our setting to Uncanny Valley to explore and understand the preferences of viewers based on cognitive dissonance related to whether the avatar is an actual human. 2 - Reward-Based Crowdfunding: Competition and Spillover Maryam Rahmani Moghaddam The spillover effects in crowdfunding campaigns have been identified as important issues in the literature. Blockbuster campaigns attract backers to their own category (positive effect) and drive away from the campaigns of other categories (negative effect). However, there is a lack of literature on the impact of concurrent campaigns (i.e., competition) on the outcome of the focal campaign. Contributing to the spillover and competition literature in the crowdfunding context, we introduce two new types of spillovers, namely, demand revelation spillover, ideation spillover. Using all US-based campaigns launched between February 2009 and April 2020 in the product design category on Kickstarter, we examine how a crowdfunding campaign is affected by the joint effects of (1) multiple spillovers and (2) competitions from both similar and different (type of products) campaigns. While we found significant impacts of blockbusters as in the received literature blockbusters positively (negatively) affect upcoming campaigns that sell similar (different) products these results failed our robustness checks. Our results show that while concurrent competitors negatively affect the success probability of the focal campaign, the negative effect of competitors selling similar products (within-cluster competition) is about 20 times larger than that of competitors selling different products (outside-cluster competition) due to the private good nature of these products. Moreover, our results indicate that campaigns that enjoy more R&D spillover are more likely to meet their funding goals. That is, a higher success rate is observed for campaigns launched in areas with more research and development activities. Furthermore, market demand plays a major role in determining the campaign’s outcome, and one standard deviation increase in the demand size improves the campaign’s chances of success by 8.67%. In addition, creators whose ideas are inspired by previous similar successful campaigns have significantly higher chances of success (ideation spillover). All these effects exert nonnegligible influences in crowdfunding campaigns after accounting for their measurements in our study. Lastly, we conclude this paper with implications and future directions.

126