Sentiment Analysis on Tokopedia Product Reviews using Natural Language Processing

Abstract

E-commerce is one of the shopping places that is currently widely accessed and used, especially in Indonesia. It acts as an online marketplace, allowing users to purchase or sell products and provide reviews for the purchased products. Those product reviews are crucial in determining the product quality and the user's overall satisfaction with the shop. The reviews of a product can determine whether a buyer will continue to purchase or abandon it. Sentiment analysis can be implemented to determine whether product reviews are positive or negative. Then, online store owners can use the results to evaluate their services. Sentiment analysis can be done automatically using artificial intelligence models, i.e., machine learning or deep learning. This research aims to get the best machine learning and deep learning models to perform sentiment analysis on Indonesian language product review datasets by combining models and the resulting performance comparisons. The authors proposed several natural language processing (NLP) models using TF- IDF and Word2Vec with logistic regression and support vector classification (SVC) classifiers. This research also implemented IndoBERT's pre-trained model to check its performance on sentiment analysis. Performance indicators used in this research are accuracy, precision, recall, and F1-Score. While the machine learning model is behind the deep learning model in accuracy, recall, and F1-Score. It has the highest precision performance.