Informs Annual Meeting 2017

MB27

INFORMS Houston – 2017

MB28

3 - The Measurement of Revenue Inefficiency Over Time: An Additive Perspective

350D Correlated and Interdependent Valuations in Mechanism Design Invited: Auctions Invited Session Chair: Michael Albert, Duke University, Durham, NC, malbert@cs.duke.edu 1 - The Value of Side Information for Revenue Extraction Hu Fu, University of British Columbia, Vancouver, BC, Canada, fu.hu.thu@gmail.com Suppose we have some side information regarding a single bidder to whom we try to sell a single item, does it matter if we reveal the information to the buyer? We show that the answer is yes, and give nontrivial sufficient conditions under which the revenue loss caused by revealing the information is bounded by a small factor. We also derive implications for revenue optimal auctions with multiple correlated bidders, via a Bayesian extension of Ronen’s lookahead auction. 2 - Optimal Procurement with Quality Concerns We characterize optimal procurement mechanisms in environments where each supplier’s privately known type affects both its production cost and the buyer’s willingness to pay for its product. We restrict attention to mechanisms that satisfy both incentive compatibility and individual rationality ex post. We show that when the affiliation among the suppliers’ types is low, it is optimal for the buyer to randomly select the winner if its willingness to pay is sufficiently large, and not trading at all otherwise. If instead the affiliation is high, a standard second-price auction, possibly with a reserve price, is optimal. 3 - Mechanism Design with Correlated Valuations: How Much Correlation is Necessary? In a setting where bidders’ values are correlated, an auction designer can extract the full social surplus as revenue. However, this result strongly relies on the assumption of a common prior distribution between the mechanism designer and the bidders. We show that if a bidder’s distribution is one of a countably infinite sequence of potential distributions that converges to an independent private values distribution, then there is no mechanism that can guarantee revenue more than greater than the optimal mechanism over the independent private value mechanism, even with sampling. This suggests that a high degree of correlation is essential in the application of correlated mechanism design. 350E Machine Learning in Physiological Data Analytics Sponsored: Artificial Intelligence Sponsored Session Chair: Hong Lin, linh@uhd.edu 1 - An ACT-R Model for Blog Users’ Negative Emotion Generation in Crises Jingjing Zhang, Nanjing University of Science & Technology, Jiangsu, China, zjingjing1202@163.com, Peng Wu, Yao Cai The purpose of this paper is to study the Blog users’ negative emotion generation in crises (Blog users’-NEGC). We develop an Adaptive Control of Thought Rational (ACT-R) model to represent the Blog users’-NEGC (the ACT-R-NEGC model). This model is based on two cognitive facts: memories have emotional effect restricted to the contextual part of activation, goals and confidence change over crises. An experimental study was conducted to test the efficiency of this model by using data retrieved from Sina Weibo and Twitter on food safety crises. This model strongly supported the Blog users’-NEGC, and this paper shows a way to model emotion within cognitive architecture in the context of real life crises. 2 - Understanding Reasons for Medication Nonadherence: An Exploration in Social Media using Sentiment-enriched Deep Learning Approach Jiaheng Xie, University of Arizona, 1130 E. Helen Street, Tucson, AZ, 85721, United States, xiej@email.arizona.edu, Xiao Liu Medication nonadherence (MNA) refers to the behavior when patients do not fill prescriptions at pharmacies. To take proactive measures, the stakeholders need to understand patients’ reasons for MNA. Current studies attempt to provide one- size-fits-all solutions to the “average patients.” To address this issue, we develop a semantically enhanced deep learning approach to detecting patient and drug- specific reasons for MNA using health social media data. Our model reached a precision of 86.43%, a recall of 92.53%, and an F1-score of 89.38%. This study contributes to information systems and health IT by designing a deep-learning- based framework for detecting tailored reasons for MNA. Giuseppe Lopomo, Duke University, FSB, P.O. Box 90120, Durham, NC, 27708, United States, glopomo@duke.edu Michael Albert, Duke University, 308 Research Drive, Durham, NC, 27708, United States, malbert@cs.duke.edu MB29

Samah Jradi, samah.jradi@kedgebs.com, Tatiana Bouzdine Chameeva, Juan Aparicio

In this paper, we introduce a Luenberger-type indicator based on the weighted additive distance function defined in [Aparicio et al. 2016] to measure revenue inefficiency change over time. The weighted additive distance function endows the well-known weighted additive model in data envelopment analysis (DEA) with a structure of distance function that allows to determine values related to the performance of points located in both the interior and exterior of the reference production possibility set. We consider an application to French wine production.

MB27

350C Leaders on Twitter Invited: Social Media Analytics Invited Session Chair: Sung Won Kim, University of Illinois at Urbana Champaign, Champaign, IL, 61820, United States, swk@illinois.edu 1 - Leaders and Lemmings on Twitter: Examining the Role of Cognitive and Heuristic Persuasion William Obenauer, Rensselaer Polytechnic Institute, Troy, NY, United States, Obenaw@rpi.edu Social networking sites (SNS) are a rapidly growing global force of influence. SNS are unique in that they restrict the use of communication tools such as body language and voice inflection, while giving users unfettered access to their audience, thus allowing for new types of leaders to emerge. We examine this dynamic by collecting data during a 14-day period of the 2014 Indian Parliamentary elections using Twitter’s streaming API 1.1 with a keyword filter limiting our data capture to Tweets relevant to the election. Our final data set includes 3.87 million Tweets which we analyze to identify behavioral trends that result in the emergence of leaders within SNS. 2 - Recruiting Members of Underrepresented Groups: What Women-in-Tech Meetups do Differently Jessica Clark, University of Maryland Smith School of Business, College Park, MD, 20742, United States, jclark@stern.nyu.edu Recruiting members of underrepresented groups into technology jobs is a major challenge for organizations. The goal of this work is to understand how various aspects of such organizations’ public faces affect their ability to attract diverse membership. Using data gathered from meetup.com, we compare general technology-themed Meetups to those explicitly targeting female members. We characterize members’ propensity to join and become active members of each Meetup using its textual description, events, reviews, and discussion board communications, among other aspects. 3 - The Impact of Short-term Introductory Incentives on Newmembers: Evidence from Online Health Communities Wei Chen, University of Arizona, 1130 East Helen Street, McClelland Hall 430, Tucson, AZ, 85721, United States, weichen@email.arizona.edu, Xiaofei Zhang, Bin Gu, Xitong Guo Many online communities deploy short-term extra incentive policy to motivate user contributions, especially for the new members. However, whether the communities can benefit from this policy is unclear. We draw on a policy change in a leading online health platform to answer this question. We find that even though the short-term extra incentives increase contribution quantity during the policy window, but decrease contribution quantity significantly after the policy window. The policy also does harm to the platform by decreasing contribution quality, but can help to increase user survival. Moreover, the impacts of the short- term extra incentives are contingent on users’ professional titles. 4 - How Corporate Sponsors can Shift Donations in Philanthropic Crowdfunding Markets Lauren Rhue, Wake Forest University, 1834 Wake Forest Rd, 212 Farrell Hall, Building 60, Winston-Salem, NC, 27106, United States, rhuela@wfu.edu This study examines if corporate sponsorship affects donations to disadvantaged recipients and if corporate sponsorship “crowds-in” or “crowds-out” individual donations on a philanthropic crowdfunding platform. Using data from DonorsChoose and a matched sample technique, I find that corporate sponsorship can “crowd-out” individual donors but “crowd-in” local donations to the highest poverty recipients and mitigate the conflict between “home bias” and donations to highest need recipients. There are managerial implications for platforms and corporate social programs as well as societal implications based on the potential of philanthropic platforms to reduce poverty and/or inequality.

174