Machine Learning and Transformer-based Model for Sentiment Analysis of Indonesian E-Commerce Reviews

Abstract

The growth of e-commerce in Indonesia has produced a large volume of user-generated reviews, which contain valuable knowledge for business decisions. However, analyzing this unstructured text data manually is inefficient. The purpose of this study is to improve the performance of sentiment classification on Indonesian e-commerce reviews using machine learning and transformer-based models. The test method is carried out using a public e-commerce review dataset. Three models are evaluated: Multinomial Naïve Bayes, Support Vector Machine (SVM), and IndoBERT. For machine learning models, text pre-processing is performed, and features are extracted using TF-IDF. For the transformer-based model, a fine-tuning approach is used. The results show that the IndoBERT model produces better classification accuracy than the other tested models. For the given dataset, this method obtains 94,1% in accuracy, outperforming both SVM (89,5%) and Multinomial Naïve Bayes (84,2%). The IndoBERT model, despite its higher computational cost, is the most effective for this classification task.