Performance Enhancement of Individual Learning Methods for Sentiment Analysis Using Ensemble Learning and Soft Voting Techniques

The rapid development of technology today makes using social media to express opinions continue to increase. Twitter is one of the social media that many people use to express their emotions and opinions regarding something, such as a brand, product, or service. By utilizing Twitter’s social media, we can retrieve and analyze every opinion from people’s tweets to see and determine their opinion on something, whether positive or negative. We can find out someone’s opinion based on every tweet they write using the Sentiment Analysis technique, which is a method to determine whether the sentiment in a text is positive, neutral, or negative. In conducting sentiment analysis, many individual learning methods have been applied, which only use one type of machine learning algorithm to perform sentiment classification, such as Naïve Bayes, SVM, and Logistic Regression. This individual learning method itself will have shortcomings depending on each algorithm used. The ensemble learning method has been developed to solve the problem by combining several machine learning algorithms to get better performance. Therefore, this paper proposes applying ensemble learning methods in conducting sentiment analysis by combining several different machine learning, and the soft voting method will be applied to determine the final result of sentiment classification. TF-IDF method will also be used to vectorize the text data. The proposed ensemble learning and soft voting method successfully obtained an accuracy score of 92%, outperforming the individual implementations of Naïve Bayes, SVM, and Logistic Regression algorithms, which only obtained 86%, 89%, and 89%, respectively.