2016 INFORMS Annual Meeting Program

MA08

INFORMS Nashville – 2016

MA06 102A-MCC Social Media Analytics Sponsored: Data Mining Sponsored Session

2 - Mining Process Patterns Via Electronic Medical Record Audit Logs He Zhang, University of South Florida, hezhang@usf.edu Mining the process patterns in the access logs from information systems can provide useful insights for the workow patterns. One important issue in process mining is that the workow is usually highly dynamic and the access logs are noisy. We presenta framework to analyze process models with noisy data at an abstract level. We implement our approach using several months of data from a large academic medical center. Empirical results show that our framework can extract the process models effectively. 3 - Feature Selection For Quality Assessment Of Predicted Protein Structures Shokoufeh Mirzaei, California State Polytechnic University - Pomona, CA, smirzaei@cpp.edu In the context of computational biology a scoring function determines the quality of a predicted protein structure. The Goal of this paper is to find a subset of protein features that are critical in identifying native protein structures in order to develop a scoring function. In pursuit of this goal, our method of research consists of 1) identify a set of protein features suggested by the literature, 2) use a variety of feature selection methods to select the best subset of features 3) use deep learning techniques in machine learning to identify new features. The final Outcome of this research is a subset of protein features and a new scoring function which predicts the quality of protein models. 4 - Healthcare Fraud Analysis Using Sequential Data Mining Babak Zafari, Babson College, Boston, MA, United States, zafari.babak@gmail.com It is estimated that at least three percent of annual health care spending is lost to overpayments. However, the large size and complexity of the health care system make comprehensive auditing infeasible. This resulted in the use of data mining approaches to detect unusual payments. In this talk, we propose the use of pattern discovery methods to find the anamolies in the medical claims payment data. Crowd-based Innovation Invited: Business Model Innovation Invited Session Chair: Bilal Gokpinar, UCL, London, United Kingdom, b.gokpinar@ucl.ac.uk 1 - Experience Breadth And Problem-solving In Crowdsourcing Contests: An Empirical Investigation Anant Mishra, George Mason University, Fairfax, VA, 22030, United States, amishra6@gmu.edu, Nirup M Menon, Shun Ye Online crowdsourcing contests have become a popular mechanism for addressing challenging problems. In this study, we use a multi-dimensional classification scheme to represent a contestant’s breadth of experience on a crowdsourcing platform, and examine how it impacts her performance in a contest. Using detailed archival data from TopCoder, a crowdsourcing platform that hosts contests across software development problem domains (e.g., architecture, design, testing), our results demonstrate that a contestant’s breadth of experience has a Mingfeng Lin, University of Arizona, 1130 E. Helen St, Tucson, AZ, 85721, United States, mingfeng@eller.arizona.edu, Bryan Zhang Will online alternative finance help reduce the geographical imbalance of capital that the literature has long documented of the traditional finance? Are funding distributions geographically more equitable online, and are there differences across different types of crowdfunding forms? We answer these questions by leveraging detailed transactions data from multiple major crowdfunding platforms in the UK. We compare the spatial flow in online debt crowdfunding to bank lending, and online equity crowdfunding to venture capital and private equity financing. We also compare the funding flow between different types of crowdfunding. Results show some surprising and interesting patterns. 3 - The Role Of Customer Investor Involvement In Crowdfunding Success Philipp Cornelius, University College London, London, United Kingdom, philipp.cornelius.12@ucl.ac.uk, Bilal Gokpinar Entrepreneurs and organisations increasingly use crowdfunding to fund innovation projects through a large number of customer investments. The growing literature on the topic has predominantly studied crowdfunding in terms of its financing mechanism. The involvement of customers during crowdfunding, however, goes beyond the provision of capital. As investors and prospective product customers, crowdfunders want to influence crowdfunding campaigns. Can project creators benefit from this or does an increased influence of the crowd push products into too many directions and deter other customers from funding? nuanced relationship with her performance in a contest. 2 - Spatial Distribution Of Alternative Finance MA08 103A-MCC

Chair: Julie Zhang, University of Massachusetts Lowell, One University Avenue, Lowell, MA, 01854, United States, Juheng_Zhang@uml.edu 1 - Challenging The Spatial Homogeneity Of Online Reviews Theodoros Lappas, Stevens University of Technology, tlappas@stevens.edu The popularity of online reviews has led to the emergence of large review-hosting platforms that rank businesses by aggregating their reviews. The standard aggregation approach naively considers only the review’s text, stars, and date, while ignoring the reviewer’s circumstances. In this work we hypothesize that the reviewer’s role as a local or visitor has a significant effect on the content and valence of his review. Our study reveals significant differences between the two populations and provides strong evidence against spatial homogeneity. We provide a detailed analysis of our findings and discuss their implications for consumers, business owners, and review-hosting platforms. 2 - Aspect Mining For Discovering Demand-side Knowledge In Online Customer Reviews Zhilei Qiao, Virginia Tech, qzhilei@vt.edu, G. Alan Wang Online reviews provide important demand-side knowledge from customers to improve mobile app product quality. However, discovering and quantifying potential app product new feature requests and defects from large amounts of unstructured text is a nontrivial task. In this paper, we propose a Latent Domain- Side Knowledge Analysis (LDSKA) that identifies the most critical app product new features and corresponding product feature status, simultaneously. Experimental results demonstrate that our proposed model outperforms existing LDA model. Our research has significant managerial implications for app developers, app customers and app platform providers. 3 - Strategically Information Disclosure On Social Media Julie Zhang, University of Massachusetts Lowell, Juheng_Zhang@uml.edu With the prevalence of social media, more firms are using social media to disclose financial information. Unlike 10-k forms, firms have freedom to choose when and how to disclose “good” or “bad” news on social media. There are strategic behaviors of firms in information disclosure on social media. We investigate the Rostyslav Korolov, Doctoral Student, Rensselaer Polytechnic Institute, 110 Eighth street, Troy, NY, 12180, United States, korolr@rpi.edu, Di Lu, Jingjing Wang, Guangyu Zhou, Claire Bonial, Clare Voss, Lance Kaplan, William A Wallace, Jiawei Han, Heng Ji We study the possibility of predicting a social protest based on social media messaging. We suggest that the frequency of text concerning the stages in the process of mobilization may be used to predict an imminent protest. We utilized several Natural Language Processing techniques to identify mobilization in social media. Our experiments with Twitter data collected before and during the 2015 Baltimore events show a correlation over time between volume of Twitter communications related to mobilization and occurrences of protest, thereby enabling estimation of the likelihood of a protest. MA07 102B-MCC Business Process Intelligence Sponsored: Data Mining Sponsored Session Chair: Zhe Shan, University of Cincinnati, Cincinnati, OH, United States, zhe.shan@uc.edu 1 - Tree-based Models For Longitudinal Data Peng Wang, University of Cincinnati, wangp9@ucmail.uc.edu Dan Liu, Brittany Green Classification and regression tree (CART) has been broadly applied due to its simplicity of explanation, automatic variable selection, visualization and interpretation. Previous algorithms for constructing CART for longitudinal data suffer from the computational difficulties in estimation of covariance matrix at each node. We proposed to utilize the quadratic inference function (QIF) and developed a new criterion, named RSSQ, to select the best splits. The proposed approach incorporates correlation wihout estimating the correlation parameters. Therefore we could improve the efficiency of the partition results and prediction accuracy. This is joint work with Dan Liu and Brittany Green strategic information disclosure of firms on social media. 4 - On Predicting Social Protest Using Social Media

123