Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SD66

2 - Putting Prediction into Practice: The Case of Restaurant Hygiene Inspections Michael Luca, Harvard Business School, Boston, MA, United States, Edward Glaeser, Andrew Hillis, Hyunjin Kim Partnering with Yelp and the City of Boston, we run an experiment comparing an inspector-curated list of restaurants to inspect (i.e.business-as-usual) to algorithm- created lists based on predictions of which restaurants are most likely to have health code violations. We find that even simple algorithms outperform business- as-usual, identifying 50% more violations per inspection. However, one practical barrier to implementing algorithms is compliance - inspectors were less likely comply with a directive to inspect a restaurant based on the algorithm. The algorithm also has fairness implications. For example, the algorithm-based predictions were more likely to target ethnic restaurants. Mingyan Xu, Baruch Collge, New York, NY, 10075, United States Educational crowdfunding is emerging as a salient hit on online platforms. Despite its growing popularity, the antecedents of funding success are far from certain. To help the fundraisers (i.e. teachers) better understand factors affecting funding success and improve their success rate, this study provides an empirical analysis on one large U.S. educational crowdfunding platform. Specifically, it analyzes the impact of the textual features from different components of the project description on funding success and identifies the differences of the impact on different levels of the project economic needs. The implications of the findings for fundraisers have also been discussed. 4 - How User-generated Content Predict Box-office Sales for Different Movie Genres Pei-Hua Chen, National Chiao Tung University, Hsin-Chu, 300, Taiwan, Chia-Tze Chang This study explored the factors that affect two kinds of movie-goers: innovators and imitators. For Innovators, we included casts of the movies and various movie attributes to predict box-office sales on opening weekends. In addition to the variables used for innovators, we added user-generated-contents to predict sales increase for imitators. We analyzed the effects of movie-goer emotions and experiences on box-office sales for different movie genres. The results showed that different movie genres provide different experiences and emotions for movie-goers. 5 - Using Text Mining to Analyze Consumer Brands Sentiments of Smart Watches Amarpreet Kohli, Associate Professor, University of Southern Maine, P.O.Box 9300, Portland, ME, 04104, United States, Solomon Nkhalamba, Zhenning Xu Social media has been a significant part of many businesses and organizations. Many firms are utilizing social media platforms to interact with their customers and clients to gauge value of their products and services through diverse stakeholders. The advent of social media platforms such as Twitter and Facebook have provided companies with easier access to collect customer opinions or reviews than the traditional survey methods and focus group approaches. To demonstrate the great potential of social media in unlocking the useful knowledge of market products, this paper uses text mining to analyze consumers’ twitter sentiments of smart watches. 3 - Text Analysis for Educational Crowdfunding Success: Comparison between Different Textual Components Joint Session AI/Practice Curated: Health Analytics with Online Data Sponsored: Artificial Intelligence Sponsored Session Chair: Kang Zhao, University of Iowa, Iowa City, IA, 52242, United States 1 - The Effects of Online Social Interactions on the Emotions of Depression Patients Qingpeng Zhang, City University of Hong Kong, 83 Tat Chee Avenue, 6/F, Academic 1, Kowloon, 12180, Hong Kong, Jiaqi Zhou, Xin Li Social media-based online health communities (OHCs) present a new platform for patients to seek social support, particularly for depressive patients. In this talk, we will introduce our research on characterizing the effect of social interactions in OHCs. We collected the full data of a major Chinese OHC. Quantitative analyses revealed that the social interactions with other patients in OHCs could positively influence the emotion of patients with depression. n SD66 West Bldg 105A

2 - Mining User-generated Content in an Online Smoking Cessation Community to Identify Smoking Status: A Machine Learning Approach Kang Zhao, University of Iowa, S224 PBB, Iowa City, IA, 52242,

United States, Xi Wang, Amanda Graham, Sarah Cha, Michael Amato, George Papandonatos, Amy Cohn, Jennifer Pearson

Online smoking cessation communities attract hundreds of thousands of smokers each year. Content shared by users in such communities may contain important information that could enable more effective and personally tailored cessation treatment recommendations. This study demonstrates a novel approach to identify individual users’ smoking status by applying machine learning techniques to user-generated content in online cessation communities. Evaluate by data from a popular online community for smokers in U.S., our approach can improve the performance of cessation identification by 9.7%. 3 - The Impact of Doctors Joining in Expert Groups on Individual Performance in Online Health Communities Wanxin Qiao, Beijing Institute of Technology, No.5,Zhongguancun Street, Haidian District, Beijing, 100081, China, Lini Kuang, Zhijun Yan, Tianmei Wang, Baowen Sun Despite a growing literature on the impacts of person-group fit on group performance, there is little evidence on how the person-group fit affects individual performance. Using a unique panel dataset , we explore the impact of doctor-expert group fit on doctor’s performance, and how the effects differ among doctors with different professional titles. Our results show that doctor joining in groups has positively affect individual performance, and doctors with low titles have positive impact on individual performance. Our work contributes to provide a deeper understanding of the relationship between doctors and expert groups in online health community. 4 - AI Enhanced Innovations for Large National Healthcare Survey Data Analytics Steven B. Cohen, RTI International, 701 13th Street NW, Washington, DC, 20005-3967, United States This presentation focuses on the development and implementation of AI and machine learning enhanced applications to imputation for national health and health care survey efforts that achieve efficiencies in terms of cost and time while satisfying well defined levels of accuracy that ensure data integrity. Attention is given to enhanced processes that serve as an alternative solution to manual, repetitive or time-intensive tasks; operationalize decisions based upon predefined outcome preferences and upon access to input data that sufficiently informs the decisions; and facilitate real-time interpretation and interactions for accessing and acting upon the AI-derived decisions. AI and Smart Technologies Sponsored: Information Systems Sponsored Session Chair: Xinxin Li, University of Connecticut, University of Connecticut, Storrs, CT, 06269, United States 1 - Reinforcement Mechanism Design, with Applications to Dynamic Pricing in Sponsored Search Auctions Michael Zhang, Professor, Chinese University of Hong Kong, Decision Sciences and Managerial Economics, 9th Floor, CYT Building, Shatin, N.T., Hong Kong, Weiran Shen, Binghui Peng, Hanpeng Liu, Ruohan Qian, Yan Hong, Zhi Guo, Zongyao Ding, Pengjun Lu, Pingzhong Tang We apply reinforcement learning techniques and propose what we call reinforcement mechanism design to tackle the dynamic pricing problem in sponsored search auctions. In contrast to previous works that rely on rationality and common knowledge among the bidders, we take a data-driven approach. We implement our proposed technique at a major search engine: We first train a buyer behavior model with a real bidding data set. We then put forward a reinforcement/MDP (Markov Decision Process)-based algorithm that optimizes reserve prices over time, in a GSP-like auction. Experiments demonstrate that our framework outperforms several strategies currently in use. n SD67 West Bldg 105B Joint Session ISS/Practice Curated:

114