Nsl-bp: A meta classifier model based prediction of Amazon product reviews

Abstract

In machine learning, the product rating prediction based on the semantic analysis of the consumers' reviews is a relevant topic. Amazon is one of the most popular online retailers, with millions of customers purchasing and reviewing products. In the literature, many research projects work on the rating prediction of a given review. In this research project, we introduce a novel approach to enhance the accuracy of rating prediction by machine learning methods by processing the reviewed text. We trained our model by using many methods, so we propose a combined model to predict the ratings of products corresponding to a given review content. First, using k-means and LDA, we cluster the products and topics so that it will be easy to predict the ratings having the same kind of products and reviews together. We trained low, neutral, and high models based on clusters and topics of products. Then, by adopting a stacking ensemble model, we combine Naïve Bayes, Logistic Regression, and SVM to predict the ratings. We will combine these models into a two-level stack. We called this newly introduced model, NSL model, and compared the prediction performance with other methods at state of the art.