Enhancing the Understanding of E-commerce Reviews through Aspect Extraction Techniques: A BERT-Based Approach

Abstract

The growth of online customer reviews on e-commerce platforms has led to an overwhelming volume and variety of data, making manual analysis impractical for both consumers and managers. Consequently, machine learning techniques, such as Aspect-Based Sentiment Analysis (ABSA), have gained prominence for their ability to determine sentiment at the aspect level. This study aims to fine-tune natural language processing models for aspect extraction in e-commerce customer reviews. We manually annotated 2781 online user review sentences in English and employed different extensions of the BERT model to identify implicit and explicit aspects. This approach diverges from prior studies, as our dataset comprises real user reviews from five prominent e-commerce platforms. The findings demonstrate the models’ effectiveness in extracting aspects from diverse e-commerce user reviews, yielding a deeper understanding of user-generated content and customer satisfaction trends, and providing valuable insights for managerial decision-making. This study contributes to the ABSA literature and offers practical implications for e-commerce platforms aiming to improve their products and services based on customer feedback.