Deep Learning for Sentiment Analysis in Indonesian Novel Review

Abstract

The rapid development of technology, especially in the internet field, is influencing the increasing number of texts available. In recent years, there has been an increase in research on the internet or social media to find out the sentiments in the review text. Sentiment analysis is a part of Natural Language Processing (NLP), which can help to show whether certain opinions tend to contain positive opinions or negative opinions. In this study, three sentiment polarities were studied using an Indonesian novel review dataset. Data was classified using the Long Short-Term Memory (LSTM) approach, one of the deep learning methods. To increase success rate, we used pre-trained word embedding to represent words into vectors. The analysis was performed by comparing the word embedding model using GloVe, Word2Vec i.e. Continuous Bag of Words and Skip-gram, and FastText i.e. Continuous Bag of Words and Skip-gram. The experimental results showed that sentiment analysis using the FastText Continuous Bag of Words model reached the highest accuracy of 80% while the Word2Vec Skip-gram model had the lowest accuracy of 78.3%. So, it can be concluded that the implementation of the FastText CBOW model is accurately used as a word representation to analyze sentiments on Indonesian novel review.