Impact of Feature engineering for Improved Sentiment Analysis in Amazon Product Reviews Using K-Nearest Neighbor

Abstract

Online reviews are an important factor that encourages consumers to make purchases through e-commerce. However, it is challenging to objectively assess the sentiments expressed by actual consumers due to the prevalence of fraudulent reviews. This study focuses on sentiment analysis and seeks to uncover the best feature combinations based on review and reviewer centric approach. The results of the study show that the combination of feature Rating, VerifiedPurchase, ReviewLengths, and (CV+TF-IDF) = 1400 words with the application of KNN classification provides the best accuracy rate of 83%. The results of this study can assist consumers in making purchasing decisions and seller in increasing the value of their products and services based on the feedback provided by customers.