The Influence of Negation Handling on Sentiment Analysis in Bahasa Indonesia

Abstract

The negation handling has not been the main focus in Bahasa Indonesia sentiment analysis so that sentence's polarity that has the word negation cannot be determined automatically. This study aimed to analyze the effect of negation handling in Bahasa Indonesia using First Sentiment Word (FSW), Rest of Sentence (ROS) and Fixed Window Length (FWL) with $\mathbf{n}= 1-5$. FSW will change the sentiment polarity of the first word after negation word, RoS will change the sentiment polarity of the entire word after the negation word, and FWL will change the polarity of sentiment words that exist in the scope of its window size. Results from this study indicate that the negation handling determined the sentence polarity optimally, compared to the absence of negation handling. FSW and FWL have an average accuracy increase by 2.43%, precision by 2.38%, and recall by 2.93%. FWL with $\mathbf{n}=3$ results the best with a precision, recall and accuracy value of 71.58%, 69.62%, 73.79% respectively. The negation handling on the automatic labeling process proved to be beneficial in increasing the performance of the classification.