Statistical Modeling Approach Using Unstructured Data

Abstract

This paper aims at finding the key factors that drive the average star ratings at e-commerce site for any product. The text reviews provided by customers on the websites are analyzed. The customer sentiments scores are calculated from the text reviews using text mining and are placed into various categories. A model is built with sentiment scores of the categories as the independent variable and average star rating as the dependent. The method also finds which categories drive customers from lower star ratings to higher star ratings. Shapley Value regression is used to find major drivers driving average star rating while ordinal logistic regression is used to identify drivers that drive customers from low star ratings to higher star ratings. We have used this methodology on customer reviews about HPs Inkjet printers on Amazon.com website and the recommendations have helped HP to improve its star ratings on the website for its Inkjet products.