2016 INFORMS Annual Meeting Program

MB48

INFORMS Nashville – 2016

3 - Using Early Click Information In Online Flash Sales Campaigns Stefano Nasini, Assistant Professor, IESEG School of Management, Lille/Paris, France, s.nasini@ieseg.fr, Victor Martínez de Albéniz, Arnau Planas In online flash sales, products are heavily discounted during very short sales periods. There is significant uncertainty over product sales, that can be reduced using the chain of sequential decisions that customers take in the website. We build a statistical model based on four layers of conditional probabilities: from (1) clicks to the main webpage to (2) clicks to a particular campaign to (3) request for information of a specific product to (4) final purchase decision. We use information from clicks occurring in the first hours of a campaign to reoptimize prices. We finally test our model with real data. 4 - Offline Assortment Optimization In The Presence Of An Online Channel Srikanth Jagabathula, NYU Stern School of Business, New York, NY, United States, sjagabat@stern.nyu.edu, Daria Dzyabura Firms are increasingly selling through both offline and online channels. The offline offerings allow the customers to physically evaluate the products and, as a result, impact the demand in both channels. Given this, we address how firms should select an offline assortment to maximize profits across both channels; we call this the showcase decision problem. We introduce a new model to incorporate the impact of physical evaluation on consumer preferences. We validate this model using a conjoint study; propose algorithms, with approximation guarantees, to determine the profit/sales maximizing assortments; and demonstrate up to 40% improvement in profits on real-world data. 210-MCC Marketing Insights from Social Media Invited: Social Media Analytics Invited Session Chair: David A. Schweidel, Emory University, 1300 Clifton Road NE, Atlanta, GA, 30322, United States, dschweidel@emory.edu 1 - Is All That Twitters Gold? Effects Of Online Chatter On Stock Market Returns And Stock Market Volatility Abhishek Borah, University of Washington, abhi7@uw.edu This study uses natural language processing to extract various dimensions across different sources of Tweets and ascertain their importance. The authors evaluate the effect of Twitter on both 1) Stock Market Returns using a Multivariate Dynamic Descriptive Panel Data Model, and 2) Stock Market Volatility using a Multivariate GARCH model. The authors find that 1) Tweets predict stock market returns and volatility in stock returns 2) Sentiment is the most important dimension and spillover effects between volatility in tweets and stock returns differ in sign depending on the sentiment of tweets, and 3) Firms’ new product announcements affect tweets. 2 - Social TV, Advertising, And Sales Beth L. Fossen, Kelley School of Business, Indiana University, Bloomington, IN, United States, bfossen@indiana.edu David A Schweidel The rapid growth of social TV - defined as the integration of social media with television programming - has outpaced the field’s understanding of how marketers can extract value out of such activities. In this research, we explore the relationship between social TV, television advertising, and sales by investigating how viewer engagement with television programs and advertisements impacts online shopping behavior. This work aims to address (1) if online chatter about television advertisements spurs sales for the advertised brand and (2) whether television programs with high online social activity are more beneficial to advertisers. 3 - Modeling Latent Homophily In Large-scale Social Networks: A Markov Random Field Approach Liye Ma, University of Maryland, liyema@rhsmith.umd.edu The rapid growth of social media platforms makes large scale social network data commonplace. Inferring consumer preference and developing targeting strategies using such data, however, remain challenging. In this study, we introduce a modeling technique, Gaussian Markov Random Fields (GMRF), to model the latent homophily of connected consumers. We show that GMRF can be applied to networks of arbitrary topology, that its conditional independence property is conceptually appealing, and that model parameters have intuitive interpretations. We analyze different model configurations incorporating one or more GMRFs, and demonstrate its application using a mobile network dataset. MB48

4 - Deriving Brand Insights With Social Media Analytics David A Schweidel, Emory University, dschweidel@emory.edu Interest in social media continues to grow. While much research has focused on the use of social media as a communication platform, limited work has probed the viability of social media data as a source of marketing insights. In this research, we examine ways in which brands may benefit from the analysis of social media data. We consider two specific applications: assessing brand health and identifying shifts in online word of mouth.

MB49

211-MCC Predicting Business Outcomes Using Social Media Invited: Social Media Analytics Invited Session Chair: Youran Fu, University of Pennsylvania, Philadelphia, PA, United States, youranfu@wharton.upenn.edu 1 - The Operational Value Of Social Media Information Ruomeng Cui, Kelley School of Business, Indiana University, cuir@indiana.edu, Santiago Gallino, Antonio Moreno-Garcia, Dennis Zhang We empirically explore how social media information helps sales forecasting. Using daily sales data from an online apparel company and publicly available Facebook posts (users’ comments and likes data), we apply various machine learning methods and find a statistically significant improvement in sales forecasts. 2 - Stock Market Movement Prediction Using Disparate Data Sources: A Probabilistic Prediction Model Bin Weng, Auburn Univeristy, Auburn, AL, 36849, United States, bzw0018@auburn.edu, Hamidreza Dolatsara, Fadel Mounir Megahed The stock market prediction has attracted much attention from academia as well as business. In recent years, social media and Internet search behavior are considered as new sources that affect human’s behavior and decision-making. The purpose of this study is to develop a probabilistic model to predict short-term stock movement by comparing machine learning methods using disparate data sources. This study not only uses traditional historic market data but also the data from technical analysis, social media, and the internet. Finally, a stock prediction tool with machine learning methods incorporated has been developed for predicting the stock’s short-term movement with high accuracy. 3 - The Value Of Social Media Data In Color Trends Forecasting Youran Fu, The Wharton School, University of Pennsylvania, Philadelphia, PA, 19104, United States, youranfu@wharton.upenn.edu, Marshall L Fisher We partnered with a leading apparel retailer to investigate how to use social media data to improve fashion color trend forecasting. We find that using fine- grained Twitter data and a Google search volume index to predict product-color sales three months out can significantly reduce forecast error compared to conventional methods. 4 - Understanding The Role And Impact Of Discussions On The Quality Of User Generated Content – The Case Of Wikipedia. Srikar Velichety, University of Arizona, srikarv@email.arizona.edu We investigate the impact of discussions on the quality of peer-produced content. Using a data science approach on the complete population of English language articles in Wikipedia, we demonstrate the predictive and explanatory power of discussions in article quality. We also compare and contrast the value of discussion characteristics with the article characteristics. Our results show that discussions add value to the predictive model by increasing both the precision and recall. On the explanatory side, we find that discussions drive both edits and diverse edits leading to better quality. Implications for theory building and policy setting in peer-production environments are discussed.

164