Sentiment analysis of Indonesian police chief using multi-level ensemble model
Wahyu Fadli Satrya, Ria Aprilliyani, Emny Harna Yossy
Abstract
Sentiment analysis is a technique of analyzing text to classify its emotion into positive, negative, or neutral sentiments. The main purpose of this study is to use sentiment analysis to seek Indonesian opinions about the chief of the Indonesian National Police. The dataset obtained from Twitter contains opinions toward the police chief from 20th August until 30th August 2022. Sentiment analysis can be done by using machine learning. Unfortunately, each machine learning algorithm performs differently depending on the data. This performance led to the issue of confusion in picking the best machine learning model to perform sentiment analysis, especially for unseen data. Therefore, ensemble learning is needed to combine multiple machine learning models. Bagging-based and Boosting-based ensemble learning have different algorithms. In order to get a high-performance model, Bagging-based and Boosting-based ensemble models can be combined as one model. This work proposed a second-level ensemble model called a multi-level ensemble model for sentiment analysis. It combines Multinomial Naive Bayes, Boosting-based ensemble learning, and Bagging-based ensemble learning. The data is processed using CountVectorizer to convert text into numbers and oversampling to handle imbalanced data. The experimental results show that our proposed model has the best results in terms of accuracy, precision, and F1-score for testing set compared to individual models.