The impact of text preprocessing on the prediction of review ratings

Abstract

With the increase of e-commerce platforms and online applications, businessmen are looking to have a ratingand review system through which they can easily reveal the feelings of customers related to their products and services. Itis undeniable from the statistics that online ratings and reviews attract new customers as well as increase sales by meansof providing confidence, ratification, opinions, comparisons, merchant credibility, etc. Although considerable researchhas been devoted to the sentiment analysis for review classification, rather less attention has been paid to the textpreprocessing which is a crucial step in opinion mining especially if convenient preprocessing strategies are found out toincrease the classification accuracy. In this paper, we concentrate on the impact of simple text preprocessing decisions inorder to predict fine-grained review rating stars whereas the majority of previous work focused on the binary distinctionof positive vs. negative. Therefore, the aim of this research is to analyze preprocessing techniques and their influence,at the same time explain the interesting observations and results on the performance of a five-class–based review ratingclassifier.