Informs Annual Meeting 2017

SC23

INFORMS Houston – 2017

3 - The Vector Representation of Medical Codes using Latent Semantic Mapping Liwen Cui, Tsinghua University, Shunde Building, Room 615, Beijing, 100084, China, cuiliwen0512@126.com, Xiaolei Xie, Zuo-Jun Max Shen, Zuo-Jun Max Shen, Haibo Wang, Rui Lu Nowadays, modern machine learning techniques are widely applied in healthcare fields, a tremendous challenge faced by previous research is how to deal with the huge amount of different medical codes that represent diagnosis or surgical operations. We propose a framework that transforms medical codes to vector representations using latent semantic mapping in natural language processing. Our method significantly outperforms various traditional dimensionality reduction methods in healthcare resources prediction, which demonstrate its application potentiality in healthcare predictive analytics. 4 - Hospital Time-to-Readmission Risk Modeling with Latent Heterogeneity Suiyao Chen, University of South Florida, 5017 Patricia Court, Apt 237, Tampa, FL, 33617, United States, suiyaochen@mail.usf.edu, Nan Kong, Xuxue Sun, Hongdao Meng, Mingyang Li Hospital readmission risk modeling is of great interest to reduce preventable readmissions and advance care quality. A readmission risk model is preferable if it exhibits superior prediction performance, identifies risk factors and constructs composite metrics to evaluate multiple hospitals. To simultaneously address these features, this paper proposes the models with incorporation of level-specific latent heterogeneity, leveraging the larger scale and less fragmented administrative claims data. To demonstrate the prediction performances of the proposed models, a real case study is considered on a state-wide heart failure patient cohort. 342D Revenue Management with Choice Models Sponsored: Revenue Management & Pricing Sponsored Session Chair: Zizhuo Wang, University of Minnesota, Minneapolis, MN, 55414, United States, zwang@umn.edu 1 - Assortment Optimization under a Single Transition Model Zizhuo Wang, University of Minnesota, 1009 5th Street SE, Minneapolis, MN, 55414, United States, zwang@umn.edu, Kameng Nip, Zhenbo Wang We consider a Markov chain choice model with single transition. In this model, customers arrive at each product with a certain probability. If the arrived product is unavailable, the seller can recommend a subset of products and the customer will purchase one of the recommended products or choose not to purchase with certain transition probabilities. We study the assortment optimization problem under this model. We show that this problem is NP-Hard while proposing polynomial time algorithms for several special cases and a compact mixed integer program formulation that can solve this problem of large size. We also provide a tight performance bound for revenue-ordered assortments. 2 - Assortment Optimization under Paired Combinatorial Logit Model Heng Zhang, University of Southern California, Los Angeles, CA, 90007, United States, hengz@usc.edu, Paat Rusmevichientong, Huseyin Topaloglu We study uncapacitated and capacitated assortment optimization under the paired combinatorial logit model, where the goal is to find a set of products to maximize the expected revenue. We show that even the unconstrained problem is strongly NP-hard. We cast both problems as one of finding the fixed point of a function, whose evaluation requires solving a non-linear integer program. Relaxing this integer program provides a framework for approximation algorithms. We use LP and SDP relaxations with randomized rounding to give 0.6 and 0.79 performance guarantees for the un-capacitated problem, and use LP relaxation with iterated rounding to give a 0.25 performance ratio for the capacitated problem. 3 - Multi-stage Assortment Problems under the Multinomial Logit Model Yuhang Ma, Cornell University, Ithaca, NY, United States, ym367@cornell.edu, Huseyin Topaloglu We consider an assortment problem where we offer sets of products in multiple stages and the choice process at each stage is driven by the multinomial logit model. In particular, we have K stages. At each stage, we offer a distinct set of products. If the customer makes a purchase at a certain stage, then her choice process terminates. If the customer does not make a purchase at a certain stage, then she observes the set of products offered at the next stage. The goal is to find a set of products to offer at each stage to maximize the expected revenue obtained from a customer. We show that the problem is NP-hard and develop an FPTAS. SC22

4 - Data Driven Assortment Optimization Velibor Misic, UCLA Anderson School of Management, Los Angeles, CA, 90095, United States, velibor.misic@anderson.ucla.edu, Dimitris Bertsimas We present a procedure for transforming data into assortment decisions through a nonparametric, ranking-based choice model. Our procedure consists of an estimation step, which involves solving a large-scale linear optimization problem using column generation to obtain a ranking-based model, and an optimization step, which involves solving a mixed-integer optimization model to obtain the assortment. We provide theoretical results on the tractability of our formulations. We evaluate the effectiveness of our method numerically with real and synthetic data and show that it outperforms alternative parametric and nonparametric proposals. 342E Pricing and Revenue Management Sponsored: Revenue Management & Pricing Sponsored Session Chair: Yanqi Xu, Princess Cruises, Valencia, CA, 91355, United States, yanqi6@yahoo.com 1 - Quantile Regression - Another Option to Calculate Elasticity for Market Response Models Justin Ji, Consultant, Operations Research, Revenue Analytics, Atlanta, GA, United States, jj@revenueanalytics.com, Pratik Mital Quantile regression was first introduced by Roger Koenker and Gilbert Bassett in 1978. It offers more complete statistical model than traditional least squares regression. Quantile regression targets at estimating the conditional median or quantiles of the response variable via investigating the relative level of influences for predictor variables. In addition, quantile regression method is more robust against outliers. In this study, the authors will use traditional least square regression model and quantile regression model to calculate elasticity for market response model. 2 - Emerging Topics in Revenue Management for the Cruise Industry Greg Vogel, Holland America, Seattle, WA, United States, gvogel@hollandamerica.com Much research has been done in traditional travel and hospitality industries such as Airlines, Hotels, Resorts, Casinos etc. Very little has been done in regards to the specific problem that are cruise lines. In this talk we discuss similarities and differences between traditional OR research into these industries and the cruise industry and then delve into the problems that the industry is most interested in solving along with an overview of problems that are currently being investigated. 3 - Dynamic and Static Airline Overbooking for Network Wei Wang, Ph.D, PROS, Houston, TX, United States, weiwang@pros.com Overbooking and cancellation are important aspects of revenue management, and much research including dynamic programming-based solutions is available in the literature. However, in practice airlines today are still applying static approaches. As an extension to our previous work for single leg, we present a simulation-based comparison of different models under both dynamic and static policies on the network, in particular we allow both no-show refund and cancellations refund to be class dependent. 4 - Maximizing Impact of Customer Reviews & Ratings Sentiment Analysis for Hoteliers Maximizing Impact of Customer Reviews & Ratings Sentiment Analysis for Hoteliers Customers’ choices seemingly are limitless and so too are their outlets for sharing experiences & opinions. This leaves hotels struggling to influence the customer decision journey, protect their brand and win business. As the volume of social media content explodes, how can hotels leverage data from customers’ online activity to deliver a truly unique experience and capture greater market share? In this session, actionable steps will be offered to: 1. Select and process the right customer data from the vast amounts available. 2. Build a strategic roadmap to drive incremental ROI.3. Create greater customer loyalty. SC23 Vinodh Balaraman, ZS Associates, NJ, United States, vinodh.balaraman@zs.com

75