LOOKING TO THE STARS: A HYBRID VOTING APPROACH FOR STAR RATING PREDICTION OF AMAZON CUSTOMER REVIEWS

ECU Author/Contributor (non-ECU co-authors, if there are any, appear on document)

Storm Pierce Davis (Creator)

Institution

East Carolina University (ECU )

Web Site: http://www.ecu.edu/lib/

Abstract: Due to the continuous growth of E-Commerce platforms, more and more data is becoming widely available. Almost every online transaction, from buying products on Amazon to renting vacation houses on Airbnb, allows for users to leave feedback in the form of customer reviews. Therefore, there is a need to provide an overview of the status of customer review analysis. The first part of this thesis systematically reviews research works conducted in the past ten years (2010-2020) to identify how customer reviews are being analyzed and how this analysis can serve a purpose to the consumer or the vendor. Common machine learning algorithms and datasets will be presented as well as the numerous applications of customer review analysis. This section aims to provide insight into what work has already been done in this field as well as discusses directions for future research. The second part of this thesis proposes a novel approach to star rating classification using customer reviews. TF-IDF techniques were used to extract key features from the review text. After evaluating the performance of five baseline classifiers through 5-fold cross validation, we incorporate the use of an ensemble hard voting classifier that employs multiple baseline classifiers to aid in final star rating prediction. Experimental results using an Amazon customer review dataset demonstrated that the best performing ensemble voting model was able to perform just as well as the best performing individual classifier while employing the use of both Logistic Regression and LinearSVC.

Additional Information

Publication

Thesis

Language: English

Date: 2023

Subjects

customer reviews;star rating prediction;customer review analysis;classification;voting model

Email this document to