Ensemble Learning for Sentiment Analysis on Twitter Data Related to Covid-19 Preventions

The World Health Organization (WHO) declared the 2019 Coronavirus disease outbreak (Covid-19) as a pandemic and made it a trending topic on social media platforms, such as Facebook and Twitter. Unfortunately, news and opinions shared on social media affect people's mentality and create panic situations in society, but in the other hand, these opinions can be analyzed using sentiment analysis approach to generate knowledge and insight for the local government to monitor people reaction to the policies that have been issued to prevent the outbreak of Covid-19 virus. Therefore, this work aimed to propose an ensemble learning model that can classify the sentiment inside the people's opinions from Twitter. The ensemble model used Naïve Bayes Classifier, C4.5, and k-Nearest Neighbors as base learners with voting mechanism to generate the final decision. For learning, the ensemble model used a dataset containing 3884 clean data that was successfully downloaded using Twitter API related to Covid-19 outbreak prevention and processed using TF-IDF method. The dataset has two classes, i.e., ‘positive’ and ‘negative’ to represent the sentiment of the opinion in each data. The proposed model got 80.61% of accuracy, 79.49% of recall, and 81.20% of precision, after being evaluated using 10-fold Cross Validation. It also performed better when compared to several learning models using only single Machine Learning algorithm.