Performance Evaluation of Machine Learning Algorithms in Aspect-Based Sentiment Analysis on E-Commerce User Reviews

Aug 4, 2026

The study evaluates machine learning algorithms for Aspect-Based Sentiment Analysis (ABSA) applied to e-commerce user reviews in Indonesia. It uses a 20,000-review dataset crawled from Shopee and Tokopedia on Google Play Store, with text preprocessing, aspect and sentiment annotation, model training, and performance assessment via accuracy, precision, recall, and F1-Score. Results show Naïve Bayes at 82.5%, KNN at 84.6%, Random Forest at 87.1%, and SVM achieving the highest accuracy (89.3%) and F1-Score (88.3%). The findings support using strong text representation handling models for aspect-level classification, enabling improved recommendation, service quality, and user experience strategies.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Abstract

Introduction

E-commerce growth and user review data

Need for sentiment analysis and ABSA motivation

Performance Evaluation of Machine Learning Algorithms in Aspect-Based Sentiment Analysis on E-Commerce User Reviews Mira Maharani Pakpahan1, Muhammad Halmi Dar2*, Mila Nirmala Sari Hasibuan3 1,2,3Faculty of Science and Technology, Universitas Labuhanbatu, Sumatera Utara Indonesia *Corresponding Author: [Email: ](Email: mhd.halmidar@gmail.com)[mhd.halmidar@gmail.com](Email: mhd.halmidar@gmail.com) Abstract. The rapid growth of the e-commerce industry in Indonesia has resulted in a significant surge in the number of user reviews available on various digital platforms. These reviews contain valuable information about customer experiences related to price, product quality, service, delivery, and applications. However, the massive volume of data and its unstructured nature pose challenges in extracting relevant information. Aspect-Based Sentiment Analysis (ABSA) presents an approach that can provide deeper insights by identifying sentiment towards specific aspects within a review, rather than just the overall general sentiment. This study aims to evaluate the performance of several machine learning algorithms, namely Naïve Bayes, Support Vector Machine (SVM), Random Forest, and K-Nearest Neighbors (KNN), in implementing ABSA on e-commerce user reviews in Indonesia. The dataset used consists of 20,000 user reviews of the Shopee and Tokopedia applications obtained through a crawling process on the Google Play Store. The data is processed through several stages: text preprocessing, aspect and sentiment annotation, model training, and performance evaluation using accuracy, precision, recall, and F1-Score metrics. The evaluation results showed differences in performance among the tested algorithms. Naïve Bayes achieved an accuracy of 82.5%, KNN achieved 84. 6%, Random Forest 87. 1%, while SVM provided the best performance with an accuracy of 89.3% and an F1-Score of 88.3%. This difference in performance indicates that algorithms that are better able to handle high-dimensional text representations, such as SVM, are superior in aspect-based sentiment classification compared to other methods. Thus, this study not only provides a comprehensive overview of the effectiveness of machine learning algorithms in sentiment analysis in the e-commerce sector but also provides a practical basis for developing recommendation systems, improving customer service, and enhancing user experience strategies on digital platforms. This research is expected to serve as a reference in the application of machine learning to support the growth of the e-commerce industry in Indonesia. Keywords: Aspect-Based Sentiment Analysis (ABSA); E-Commerce; Machine Learning and Sentiment Analysis. 1. INTRODUCTION E-commerce has become one of the largest pillars of the digital economy in Indonesia, with business revenues ranking highest among other ASEAN countries at Rp. 778.8 trillion [1] . With this development, Indonesia is now the largest e-commerce market in Southeast Asia. This is in line with the strategy of the Government of the Republic of Indonesia as stated in the Asta Cita of President Prabowo Subianto and Vice President Gibran Rakabuming to encourage the creative industry by creating new sources of economic growth through the development of the digital economy. The growth of e-commerce businesses in Indonesia is inseparable from the ever-expanding user base. According to the Ministry of Trade's Data and Information Systems Center (PDSI Kemendag), the number of e-commerce users has increased by 69% in the last five years, from 38.7 million users in 2020 to 65.6 million in 2024, and is expected to continue to increase to 99.1 million users in 2029 [2] . This indicates the increasing interest in using e-commerce in the country. Various e-commerce platforms exist to make it easier for users to conduct online transactions more personally and efficiently [3], [4] . Shopee ID, Tokopedia, Lazada ID, Blibli, and Bukalapak are the largest and most popula

FAQs

What problem does the study address in e-commerce reviews?

The study addresses the challenge of extracting useful information from large volumes of unstructured customer reviews, where overall sentiment is insufficient for understanding sentiment toward specific aspects.

Which machine learning algorithms are evaluated for ABSA?

Naïve Bayes, Support Vector Machine (SVM), Random Forest, and K-Nearest Neighbors (KNN) are evaluated on aspect-based sentiment classification for Indonesian e-commerce reviews.

What performance metric shows SVM as the best model?

SVM achieves the highest accuracy (89.3%) and the highest F1-Score (88.3%) among the tested algorithms.

Aug 4, 2026

Upload by Evangeline