2016 INFORMS Annual Meeting Program

MD06

INFORMS Nashville – 2016

MD06 102A-MCC Business Analytics and Text Mining Sponsored: Data Mining Sponsored Session Chair: Xiao Liu, University of Arizona, 1130 E. Helen St., Tucson, AZ, 85721, United States, xiaoliu@email.arizona.edu 1 - A Network-based Inference Model For Estimating Missing Attributes Da Xu, University of Utah, da.xu@eccles.utah.edu The attribute information, which is critical for recommendations, search engines, and advertising targeting, is valuable for business intelligence. While, all the aforementioned machine learning based applications are only capable of optimal performance when the data utilized are of high quality. And in many cases, the attribute information is incomplete, which makes a big obstacle in targeting products, businesses, or promotions effectively. In this paper, we utilized both networks and content information to infer missing business attributes, which could benefit business recommendations, help validate online business information, and provide better personalized offerings. 2 - Extracting Signals From Social Media Text With Natural Language Processing, Machine Learning And Domain Adaptation Wenli Zhang, University of Arizona, Tucson, AZ, United States, wenlizhang@email.arizona.edu, Sudha Ram There has been increasing interest in using social media data for predictive analytics in different domains. Although significant promise has been shown, mounting evidence suggests many of the results can be misrepresented because of the loosely structured text and noise caused by media spikes and use of misleading phases. We introduce efficient techniques combining Natural Language Processing and Machine Learning to extract signal from social media text. Sophisticated domain adaptation method is introduced to address multi- domain adaptation problem. The methodology can be used for extracting signals in health care and other domains with a view to enabling improved predictions. 3 - The Effect Of Rating System Design On Emotion Sharing Ying Liu, Arizona State University, yingliu_is@asu.edu How do ratings and reviews reflect consumers’ overall evaluations toward the product? Does the overall evaluation reflect average experience or is it biased? In this study, we focus on evaluating the integration bias in consumers’ rating behavior through rating system design. Analysis of data from two leading restaurant review websites with different rating systems suggests that the overall ratings tend to reflect consumers’ extreme experiences in a single-dimensional rating system, however, their average experience by taking all dimensions into consideration in multi-dimensional rating systems. The results are confirmed by information from text reviews through text mining skills. 4 - Webcasting Game Or Sharing Experience? Exploring The Role Of Team-created Word-of-mouth In Football Game Attendance Yang Wang, University of Utah, yang.wang@eccles.utah.edu, Nick Sullivan, Shyam Gopinath A 2015 NCAA report shows that college football attendance drops to the lowest in 15 years. To help generate demand, the 128 FBS teams develop different strategies and use social media as a tool to attract fans. Among them, those schools with top game attendance usually tweet a lot about game ambiance which shows the unique game experience at the stadium, while the others only webcast the team performance on the pitch. This study aims to examine the differential impacts of the two types of team-created word-of-mouth on the future game attendance versus the TV viewership. We find the unique role of each type of the content and provide relevant business implications. MD07 102B-MCC Urban Data Analytics and Mining Sponsored: Data Mining Sponsored Session Chair: Xun Zhou, University of Iowa, S210 PBB 21 East Market Street, Iowa City, IA, 52242, United States, xun-zhou@uiowa.edu 1 - A Traffic Flow Approach To Early Detection Of Gathering Events Amin Vahedian, University of Iowa, Iowa City, IA, United States, amin-vahediankhezerlou@uiowa.edu, Xun Zhou Given traffic flows in a spatial field, early detection of gathering events problem aims to discover the most likely gathering events. It is important for city planners to identify emerging gathering events which might cause public safety or sustainability issues. Here, we model the footprint of a gathering event as a

directed acyclic Graph, which captures routes of the flows to an event and their most likely destination. We also propose an efficient algorithm to discover the most likely events. Our analysis shows that the proposed model and algorithm efficiently and effectively capture important gathering events from real-world

mobility data while saving 50% time over the baseline algorithm. 2 - Mapping The Structure Of China’S Cities Network

Xiaolong Xue, Harbin Institute of Technology, xlxue@hit.edu.cn The structure of China’s cities network is dramatical changing with the rapid urbanization process. This paper analyzes the research status of cities network theory, and constructs China’s cities network model using China’s transportation infrastructure data. The structure of China’s cities network is described through network characteristics, and China’s cities network is divided into different network communities by clustering analysis. We find the center city, traffic hub and regional centers by calculating cities nodes’ effectiveness. The calculating of network effectiveness provides a reference for improving the efficiency of China’s cities network. 3 - A Markov Decision Process Approach To Optimizing Taxi Driver Business Efficiency Xun Zhou, University of Iowa, xun-zhou@uiowa.edu Improving taxi business efficiency is an important societal problem. This work investigates how to increase the revenue efficiency (revenue per unit time) of taxi drivers. To solve this problem we model the passenger seeking process as a Markov Decision Process(MDP) and learn necessary parameters from historical taxi data. A case study and several experimental evaluation on a real dataset from a major city in China show that our proposed approach improve the revenue efficiency of inexperienced drivers by up to 15%. 4 - Operation Strategies And Algorithms For Minibus Systems In Hong Kong In Hong Kong, the spatial distribution of Minibuses within the public transportation system is self-organized, lacking a clearly defined operation strategy. There is no optimization based on current demand. Within this paper several operation strategies are introduced. A new integrated algorithm for optimal strategy is described in detail, including two approaches: a user-based approach, outlining a strategy to capture and optimize consumer demand, and an operation-based approach, outlining a strategy to balance revenue and consumer satisfaction. MD08 103A-MCC Tutorial: Data-Driven Research in Revenue Management Invited: Business Model Innovation Invited Session Chair: David Simchi-Levi, Massachusetts Institute of Technology, Masachusetts Avenue, Cambridge, MA, 0, United States, dslevi@mit.edu 1 - Data-Driven Research In Revenue Management We present a pricing optimization problem for the data plans of a big satellite firm. First we address the problem of missing data (as reservation prices are not directly observed especially for those who are not current customers). We formulate the price optimization problem as a MIP and develop properties and heuristics in order to solve realistic instances providing analytical lower bounds of their performance. We conclude that with our method the company can increase its profits by more than 10% and outperform the current plans’ prices even under misspecifications of the assumptions. Jacky Pak Ki Li, PhD Student, VU University Amsterdam, De Boelelaan 1081a, Amsterdam, 1081 HV, Netherlands, jacky.li@kpu.ca David Simchi-Levi, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA, United States, dslevi@mit.edu

210