Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

WD66

n WD64 West Bldg 104A Joint Session DM/Practice Curated: Data Science for Text Analysis Sponsored: Data Mining Sponsored Session Chair: Babak Zafari, Babson College, Babson Hall 218C, Babson Park, MA, 02457, United States 1 - Text Mining Analysis of Business Data Analytics and Data Science Jobs Requirements Zinovy Radovilsky, Professor, California State University-East Bay, 25800 Carlos Bee Boulevard, Hayward, CA, 94506, United States, Vishwanath Hegde, Anuja Acharya, Uma Uma We identify and compare knowledge and skills for business data analytics (BDA) and data science (DS) professions. We collected primary BDA and DS job posting data from online job-related websites, developed document data matrix, and applied text mining analysis including singular vector decomposition, VARIMAX rotation, and latent class analysis. Based on this text mining analysis, we identified main similarities of and important differences between the BDA and DS job requirements. These results provide vital insights for designing curriculum and training in the evolving BDA and DS areas, and also enable professional to sharpen their skills aligned with job market requirements. 2 - Deception or Truth? The Impact of Linguistic Cues to Fraud on Capital-giving Willingness: Evidence from Crowdfunding Market Xicheng Yin, Tongji University, Shanghai, China, Wei Wang, Kevin Zhu, Hongwei Wang, Pei Yin, Wei Chen We adopt the following indicators to measure the linguistic cues to fraud: Cognitive Load, Internal Imagination, Dissociation, Negative Emotion, Lexical Diversity, Lexical Ease of Read, Lexical Complexity and New Word Ratio as well. Locational prepositions and temporal prepositions in reward statement, non-first- person prepositions in blurb and reward statement, and cohesion in project description normally result in successful campaigns. The concreteness of blurb and detailed description makes the project look like a deception. We provide a method to detect linguistic cues to fraud on crowdfunding campaigns and provide suggestions to project founders to better describe their projects. 3 - Knowledge Mining in Scientific Literature for Complex Social Problems: An Example using Multi-stakeholder Performance Management Victor Zitian Chen, University of North Carolina, Charlotte, NC, United States, Reza Mousavi, Wlodek Zadrozny The volume of scientific publications and the degree of knowledge fragmentation creates information overload problems and makes knowledge synthesis for solving complex social problems exceedingly difficult. Built on text mining techniques and keyword dictionaries, we develop a literature review algorithm to automate the search, comparing, and grouping of predictors of corporate performance measures from multiple stakeholders’ perspectives (customers, employees, societies, and investors). 4 - Expanding a Theoretical Model with Survey Simulations Seyede Yasaman Amirkiaee, PhD Candidate and Teaching Fellow, University of North Texas, 1307 West Highland Street, BLB 357D, Denton, TX, 76201, United States, Nicholas Evangelopoulos Traditional survey design involves iterations of instrument development steps that include assessments of item reliability, and construct convergent and discriminant validity. These iterations are time consuming and tend to overuse human subjects. In an effort to economize on these resources, this research uses traditional survey methods to fit a theoretical model for the intention to ride a self-driving vehicle, and then expands the model to include additional constructs by performing survey simulations that do not involve human participants. 5 - Topic Modelling for Medical Prescription Fraud and Abuse Detection Babak Zafari, Babson College, Babson Hall 218C, Babson Park, MA, 02457, United States, Tahir Ekin Medical prescription fraud and abuse has been a pressing issue in the U.S. resulting in large financial losses and adverse effects on human health. In this work, we use the real world Medicare Part D prescription data to analyze prescriber-drug associations. In particular, we propose the use of topic models to group drugs with respect to the billing patterns and exhibit the potential aberrant behaviors. The prescription patterns of the providers are retrieved with an emphasis on opioids, and aggregated into distance based measures which are visualized by concentration functions. This output can enable medical auditors to identify leads for audits of medical providers.

n WD66 West Bldg 105A Practice - Quality Management & Applications Contributed Session Chair: Bhupesh Shetty, University of Iowa, Iowa City, IA, 07030, United States 1 - Real-time Data-driven Monitoring for High Dimensional Multistage Manufacturing Process Mohammadhossein Amini, PhD Candidate, Kansas State University, 2061 Rathbone Hall 1701B Platt St, Manhattan, KS, 66506, United States Today’s manufacturing processes are much more complex including multiple stages, and sensors are embedded throughout the processes that generate a huge amount of data in high dimensions. However, current quality engineering practices still do not efficiently use the generated data. This study aims to develop predictive models using machines production data to monitor the high dimensional multistage process in real-time. 2 - Dual Response Surface Optimization of One Side Single Quality Dual Response Surface methodology is a powerful tool for simultaneously optimizing the mean and the variance of a quality characteristic in the field of quality engineering. The optimization of dual response systems in order to achieve better quality has played a major role in the design of industrial products and processes. In this work we derive a new method to optimize the mean and variance of a product or process defined by a single quality characteristic and has a one sided specification limit. For this aim, we suggest the usage of the process capability index as the objective function to find the optimal values of the control variables thus allowing to maximize the yield of the process. 3 - An Image Based Multivariate EWMA Control Chart with PCA for Faults Detection under Multiple Capturing Settings Shengfeng Chen, Western Michigan University, Kalamazoo, MI, 49008, United States, Lee Wells Machine vision systems are increasingly being used in manufacturing shop floor in product quality control. Our approach extends the current research that use only one image capturing setting for fault detection. As image quality can be affected by light intensity, color and glare, we propose multiple capturing settings for better fault detection. In addition, the use of a multivariate EWMA for monitoring industrial products in conjunction with PCA analysis are used to deal with the increased data dimensions. Extensive computer simulations show that the proposed multiple capturing settings outperform the single capturing setting with greater in-control and out-of-control ratios. 4 - Identifying Sources of Assignable Error via Process Pattern Mining Bhupesh Shetty, PhD Student, University of Iowa, Iowa City, IA, 52245, United States, Nick Street, Jeffrey W. Ohlmann We explore the problem of identifying the root cause of product defects using event logs in a manufacturing process. We use a pattern mining algorithm based on Apriori to identify frequent patterns, and use binary correlation measures to identify patterns associated with elevated error rate. We design a simulation model to generate synthetic datasets to test our algorithm. We compare the effectiveness of different correlation measures, target pattern complexities, and sample sizes with and without knowledge of the underlying process. We show that knowledge of the underlying process helps in identifying the pattern that is associated with defects. Characteristic Using the CPK Process Indicator Index Michael Bendersky, Ben Gurion University of the Negev, Beersheba, Israel

515