Sentiment Analysis Based on Tweet Reply at Public Figure Account using Machine Learning and Latent Semantic Analysis

Abstract

Twitter is a social media platform that enables its user to communicate or see various running events online. It also enables everyone to share information. Public figures, such as celebrities, artists, or politicians often dominate the talk and trending topics. The pros and cons among the public figure accounts cause either positive or negative sentiments. Thus this condition urges a system that can classify each of the user's replies to a public figure account as a consideration to change to a better communication pattern. Some of the possible methods are the naive Bayes, SVM, and logistic regression, all of these methods are combined with the Latent Semantic Analysis (LSA). The classification system will be based on 1500 dataset that has been labeled and divided into 80% training data and 20% testing data. The result of the confusion matrix showed the highest accuracy for SVM 80.4%, logistic regression 80.6%, and multinomial Naive Bayes 78.6%.