This Text Has the Scent of Starbucks: A Laplacian Structured Sparsity Model for Computational Branding Analytics

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1325–1336, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics This Text Has the Scent of Starbucks: A Laplacian Structured Sparsity Model for Computational Branding Analytics William Yang Wang School of Computer Science Carnegie Mellon University ww@cmu.edu Edward Lin and John Kominek Voci Technologies, Inc. Pittsburgh, PA 15217 {ed.lin,john.kominek}@vocitec.com Abstract We propose a Laplacian structured sparsity model to study computational branding ana- lytics. To do this, we collected customer re- views from Starbucks, Dunkin’ Donuts, and other coffee shops across 38 major cities in the Midwest and Northeastern regions of USA. We study the brand related language use through these reviews, with focuses on the brand satisfaction and gender factors. In particular, we perform three tasks: auto- matic brand identification from raw text, joint brand-satisfaction prediction, and joint brand- gender-satisfaction prediction. This work ex- tends previous studies in text classification by incorporating the dependency and interaction among local features in the form of structured sparsity in a log-linear model. Our quantita- tive evaluation shows that our approach which combines the advantages of graphical model- ing and sparsity modeling techniques signifi- cantly outperforms various standard and state- of-the-art text classification algorithms. In ad- dition, qualitative analysis of our model re- veals important features of the language uses associated with the specific brands. 1 Introduction In marketing science, branding is a modern market- ing strategy of creating a unique image for a prod- uct in the customers’ mind. Establishing the brand in the broad social context is just as important as building a good product (Makens, 1965; Lederer and Hill, 2001; Kim et al., 2013). In fact, blind taste test experiments have frequently shown how branding directly leads to the success of products and companies. Most notably is a continued study sponsored by Pepsi, known as the Pepsi Challenge1, where Pepsi demonstrates how even though people preferred the taste of Pepsi, Coca-Cola’s branding has made it more popular. Even now, Microsoft uses similar blind taste tests2 to compare search en- gines, Bing and Google, showing that although par- ticipants prefer Bing’s results, Google’s brand might have strengthened over the years. These studies all suggest that brand and its associations play impor- tant roles in the customers’ perceptions and deci- sions. To accommodate the market change, companies frequently adjust branding strategies by analyzing how their customers receive and respond to brand- ing messages. So far, such analysis is often done by using surveys and focus groups (Moon and Quelch, 2006), which is expensive and not time- efficient. Recently, with the advance of machine learning techniques, researchers from the chemistry and vision communities started to pay attention to the problem of automatic brand identification from smell (Luo et al., 2004) and images (Pelisson et al., 2003). In contrast, even though textual data that contains hidden branding information is abundantly available in many forms over the Web, automatic discovery and computational analysis on such data are not well studied in …

Content truncated to 500 words