Modelling Sentiment Analysis on Indonesian Product Reviews Using Machine Learning

Abstract

Insight from product reviews from customers is important messages to the business. It provides feedback to the business and enhances the customers’ experiences and leading to sustainable competitive advantage for the business. Sentiment analysis can be performed from the customers’ product reviews regarding to particular product or service. Capturing sentiment analysis can be done automatically using machine learning techniques. This research aims to explore several machine learning algorithms to model automatic sentiment analysis on Indonesian product reviews. The product reviews data was gathered from an e-commerce platform in Indonesia. All of the product reviews are in the local (i.e. Indonesian) language. The research contributes to a state of the art of automatic sentiment analysis model on Indonesian product reviews. Eight algorithms and 4507 settings are explored to find the best setting and algorithm to model sentiment analysis in Indonesian product reviews. The results demonstrate that the best model was achieved by the one trained with five layers of artificial neural network with 99.1% for testing model accuracy, precision, recall, and F1-score. The best AUC of the model was 99.6%. The model requires 34.68 minutes to be trained.