Sentiment Analysis of Twitter Users Regarding Taxation Topics in Indonesia Utilizing Multinomial Naive Bayes

Abstract

The country's income is heavily dependent on taxes, which contribute to improved public well-being. Public confidence in tax authorities plays a key role in increasing tax receipts. Therefore, it is important to measure this level of confidence. One of the methods used is sentimental analysis, which helps to understand public views on regulations, services, performance, and tax policies. One of the purposes of this study is to measure the sentiment of Twitter users towards taxation in Indonesia. Sentiment analysis involves data collection processes, initial data processing, separation of datasets, feature extraction, classification, and evaluation. The classification model used is Multinomial Naive Bayes with a comparison of 80% training data and 20% test data. The results show that 89.65% of tweets about taxation in Indonesia have negative sentiment. The model evaluation was carried out on two test scenarios, namely initial data and randomly under-sampleed data. Classification on initial data achieved accuracy of 89.97%, precision of 46.68%, and sensitivity of 33.61%. Whereas on undersampling data results, accuration reached 53.28%, accurateness of 52.66%, and sensibility of 52.52%. Analysis showed significant differences between the two scenarios in which undersammpling techniques resulted in a more balanced distribution of data. Despite this, the model still faces difficulties in classifying positive and neutral data due to the dominance of negative sentiment.