Ensemble approach for sentiment polarity analysis in user-generated Indonesian text

Abstract

An ensemble of machine learning approach is presented to detect the sentiment polarity in Indonesian user-generated text. We use TF-IDF, term presence, and valence score of sentiment lexicon as features in learning process for classifier model generation. The valence score is resulted from a system which considers the negation and booster words as well. Our experiments with ensemble approach for sentiment polarity analysis on Indonesian user generated text slightly better than single classifier approach and outperform the baseline methods, which gives 91.59% accuracy and 91.59% F1-score.