Optimizing Customer Decision-Making with Enhanced Product Evaluation Using Aspect-Based Sentiment Analysis and Naïve Bayes Algorithm

Abstract

In the wake of the pandemic–driven surge in e-commerce usage, consumers increasingly rely on online reviews to assess product quality. However, customer feedback's sheer volume and inconsistency make it difficult to extract meaningful insights. This study aims to enhance customer decision-making by developing an online product evaluator system that utilizes Aspect-Based Sentiment Analysis (ABSA) and the Naïve Bayes algorithm to provide fine-grained sentiment insights. The system collects user reviews from prominent platforms such as Shopee, Amazon, and Lazada. Reviews undergo natural language pre-processing, including tokenization, stop-word removal, and lemmatization. Aspect terms are extracted using frequency-based techniques, and sentiment classification is performed using the Naïve Bayes algorithm. Additionally, an abstractive summarization module is implemented to provide concise summaries for each aspect. The trained sentiment classifier achieved an accuracy of 90% based on validation using a confusion matrix and classification report. The aspect-based approach delivered more detailed insights compared to traditional sentiment analysis. A usability test following the FURPS model (Functionality, Usability, Reliability, Performance, and Supportability) yields an overall rating of 4.24, interpreted as “Strongly Agree,” from a sample of 50 users, including customers, professors, IT developers, and sellers. The results indicated that the system effectively supports customers in evaluating products through targeted sentiment analysis, thereby improving the quality and efficiency of purchase decisions. Moreover, the findings validate the integration of ABSA and machine learning as a robust solution in e-commerce environments.