Consumer preference estimation from Twitter classification: Validation and uncertainty analysis
Editor: Udo Lindemann, Srinivasan V, Yong Se Kim, Sang Won Lee, John Clarkson, Gaetano Cascini
Author: Stone, Thomas Michael; Choi, Seung-Kyum
Institution: Georgia Institute of Technology, United States of America
In recent years, the membership and activity of Twitter, Facebook, blogs, and other user-generate content sites has experience significant growth. Users express their opinions regarding a wide range of topics, including consumer products and services. Thus, these sites have the potential to facilitate product design via the extraction of consumer opinion and sentiment regarding product features. A key challenge is how to appropriately extract consumer preferences from the messages. This challenge is addressed with respect to Twitter using a smartphone case study. Twitter messages regarding particular smartphone attributes are classified according to sentiment: positive, negative, or neutral. This sentiment information is then used to develop an estimate of consumer preference for particular smartphone attributes, such as battery life or screen size. Uncertainty analysis is conducted in order to assess the effects of sentiment classification accuracy. Validation techniques indicate that a revised framework would be useful for predicting consumer decisions and facilitating product design; however refinement in terms of comprehensiveness and accuracy or needed.