SEARCH

SEARCH BY CITATION

Keywords:

  • topic model;
  • information retrieval;
  • quality control;
  • unsupervised learning

This article proposes a method for Pareto charting that is based on unsupervised, freestyle text such as customer complaint, rework, scrap, or maintenance event descriptions. The proposed procedure is based on a slight extension of the latent Dirichlet allocation method to form multifield latent Dirichlet allocation. The extension is the usage of field-specific dictionaries for multifield databases and changes to recommended default prior settings. We use a numerical study to motivate the prior setting selection. A real-world case study associated with user reviews of Toyota Camry vehicles is used to illustrate the practical value of the proposed methods. The results indicate that only 4% of the words written by Consumer Reports reviewers from the last 10 years relate to the widely publicized unintended acceleration issue. Copyright © 2012 John Wiley & Sons, Ltd.