Analyzing Amazon CD Reviews with Bayesian Monitoring and Machine Learning Methods

Abstract

This paper analyzes customer reviews of CDs sold on Amazon.com using various statistical and machine learning methods. We investigated the distribution properties through exploratory analyses and the Bayesian monitoring method was utilized to analyze life cycles of CDs. We proposed an adjustment to the classic Bayesian monitoring technique which allows it to deal with extreme changes in data. To predict how many reviews CDs get, we compared the performances of a range of machine learning models and identified important features affecting the number of reviews using permutation importance.