Aspect-based sentiment classification of user reviews to understand customer satisfaction of e-commerce platforms

Abstract

Making use of user-generated content, in particular customer reviews, is an essential means for companies to gather information about customers. In order to utilize reviews in improving customer satisfaction, companies need to identify the important components of their business that are discussed by the customers. In this article, we propose to use aspect-based sentiment analysis for reviews focusing on e-commerce platforms. Fourteen aspects were extracted from the literature and empirical analysis of 3500 randomly selected reviews from the Trustpilot platform. A unique dataset was created by manually annotating the reviews and assigning a sentiment for each aspect-review pair. Aspect-based sentiment classification was performed using some of the most recent machine learning models, with RoBERTa identified as the best-performing model, achieving over 90% accuracy. We also demonstrate how insights generated from the sentiment scores can assist companies in improving their service and increasing customer satisfaction.