Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SD54

3 - Competition and Coopetition among Social Media Content in Disasters Eunae Yoo, University of Tennessee, Knoxville, TN, United States, Elliot Rabinovich, Bin Gu Humanitarian organizations may convey similar content when communicating on social media platforms during disasters. Our study examines if the diffusion for a piece of content is detracted by or is augmented by the spread of similar content. Therefore, we test whether a competitive or coopetitive dynamic exists among similar content. To that end, we formulate a generalized Hawkes model and estimate the model using Twitter data from multiple disasters. The results from our analysis indicate that similar content tends to interact competitively. Based on the results, we provide implications for humanitarian organizations for coordinating the release of similar social media content during crises. 4 - Reducing Fear through Credible Information During Crisis Tricia Moravec, Indiana University, Bloomington, IN, United States The increase in global natural disasters and crises throughout the past 10 years, combined with the increased usage of online social media in transmitting information during these disasters, had led to increased need to understand the process through which social media enables at-risk people to stay informed. People panic in the face of a natural disaster, and the first method to enable rational response is to calm the public down. We examine whether information provided from one with an authoritative, reliable reputation helps people calm down, so they may more rationally deal with the crisis. We find that increased informational support does negatively influence fear, such that fear decreases when information is posted by the premier authority on wildfire fighting. Our results suggest that informational support is highly effective in enabling people to make the most effective decisions during the disaster response phase by helping them toward a more rationale state of mind. Chair: Sven Seuken, University of Zurich, Zurich, 8050, Switzerland Co-Chair: Benjamin Lubin, Boston University, Boston, MA, 02215, United States 1 - The Power of Machine Learning and Market Design for Cloud Computing Admission Control Ludwig Dierks, University of Zurich, Zurich, Switzerland, Ian Kash, Sven Seuken Cloud computing providers must handle customer workloads that wish to scale their use of resources such as virtual machines up and down over time. Currently, this is often done using simple threshold policies to reserve large parts of compute clusters, which leads to a low average utilization of the cluster. We present new, more sophisticated policies that can take learned or elicited probabilistic information about the future behaviors of arriving workloads into account. Based on these policies, we show the high potential machine learning and market design have for increasing efficiency. 2 - Combinatorial Auctions via Machine Learning-based Preference Elicitation Gianluca Brero, University of Zurich, Stettbachstrasse, 93, Zurich, 8051, Switzerland, Benjamin Lubin, Sven Seuken Combinatorial auctions (CAs) are used to allocate items among bidders with complex valuations. Since the value space grows exponentially in the number of items, it is often impossible for bidders to report all their values, thus leading to inefficiencies. We introduce a machine learning-based elicitation algorithm to identify which values to query from the bidders. We integrate this algorithm in a new CA mechanism we call PVM where payments are determined so that bidders’ incentives are aligned with allocative efficiency. We validate PVM experimentally and we show that it achieves high allocative efficiency even when only few values are elicited from the bidders. 3 - A Bayesian Clearing Mechanism for Combinatorial Auctions S. Bastien Lahaie, Google, New York, NY, 10011, United States, Gianluca Brero We cast the problem of combinatorial auction design in a Bayesian framework to incorporate prior information into the auction process and minimize the number of rounds. We develop a generative model of bidder valuations and market prices, which then forms the basis of an auction process which alternates between refining estimates of bidder valuations, and computing candidate clearing prices. We provide an implementation of the auction, empirically evaluate it over a range of valuation domains, and find that it is extremely competitive against a conventional combinatorial clock auction. n SD53 North Bldg 232A Machine Learning in Market Design Sponsored: Auction and Marketing Design Sponsored Session

4 - Learning to Branch for Winner Determination in Combinatorial Auctions

Tuomas W. Sandholm, Carnegie Mellon University, Gates Center for Computer Science, Pittsburgh, PA, 15213, United States, Maria Florina Balcan, Travis Dick, Ellen Vitercik Tree search algorithms, e.g. branch-and-bound, are the most widely used tools for combinatorial/nonconvex problems - e.g. MIP/CSP. To get small trees, it is key to decide, when expanding a node, on which question (e.g. variable) to branch. Many techniques have been proposed, but theory was lacking. We show how to use machine learning to determine an optimal weighting of any set of branching rules for the instance distribution at hand using samples from the distribution. We prove the first sample complexity guarantees for tree search algorithm configuration. The theory gives rise to a learning algorithm. It dramatically reduces tree size, e.g. in optimal combinatorial auction winner determination. n SD54 North Bldg 232B Behavioral Research in Sustainability and Supply Chains Sponsored: Behavioral Operations Management Sponsored Session Chair: Stephen Leider, University of Michigan, Ann Arbor, MI, 48109, United States 1 - Relative Performance Transparency: Effects on Sustainable Consumer Behavior Yanchong Zheng, MIT, Cambridge, MA, 02142, United States, Ryan Buell, Shwetha P. Mariadassou We study how transparency into the levels and changes of relative sustainability performance affects consumer behavior. In a series of online consumer choice studies, we show that when the relative performance concerns comparison of a firm to its competitors, transparency into the firm’s relative performance level has a stronger impact on behavior than transparency into the firm’s relative performance change. Conversely, when concerning performance comparison of the consumer him/herself to other consumers, transparency into relative performance changes has a more dominant effect on behavior. We identify the mechanisms underlying these results and confirm them in our last study. 2 - Shared Supplier Capacity as a Barrier to Socially Responsible Sourcing Jacob Chestnut, Cornell College of Business, SHA, Ithaca, NY, United Statesu, Ravi Anupindi This experimental project considers the role of buyer behavior (e.g., time pressure, low margin, near delivery specification changes) in their supplier’s performance along the dimension of social sustainability (forced OT, child labor, unauthorized outsourcing, etc.). We attempt to understand the relevant features (contractual and non-contractual) that suppliers use when creating a preference ranking over buyers. Doing so allows us to assess the effectiveness of the operational and non-operational levers that a “good buyer might employ to improve their rank, thereby decreasing the likelihood that deviations occur on their order. 3 - The Role of Psychological Distance and Construal Level on Sustainable Operations Decision Making Saif Mir, Assistant Professor, College of Charleston, Charleston, SC, 29424, United States, Stephanie Eckerd, John Aloysius We evaluate the role of psychological distances and construal levels in influencing operations manager propensity to make sustainable decisions. Previous studies have largely investigated cases when the two mechanisms are congruent. Specifically, near psychological distances are linked with concrete construal and far psychological distances are linked with abstract construal. In our study we evaluate the interaction of these mechanisms and the resulting effect on decision making behavior, using a 2 (Psychological distance: Near/Far) x 2 (Construal level: High/Low) vignette experiment. 4 - Building Alliances for Socially Responsible Sourcing Han Zhang, Student, Kelley School of Business, Indiana University Bloomington, 1309 E. 10th St Ste 4100, Bloomington, IN, 47405, United States, Ruth Beer, Kyle D. Cattani Following the mandate by the Dodd-Frank Act that US public companies disclose the use of conflict minerals in their products and processes, five well-known companies established a fund to audit mineral suppliers. This initiative has been a success at raising funds despite the incentives of companies to free ride on the contributions by other companies. Our game-theoretical model incorporates two elements that explain the phenomenon: an invitation stage and heterogeneity in the statuses of the firms. Our lab experiment shows that the invitation stage is key to high contributions, and status significantly influences whether the invitation succeeds.

109