Classification of Public Sentiment on Fuel Price Increases Using CNN

Abstract

The government's policy of changing fuel prices is carried out every year. The public gave responses to this policy categorized as positive, negative, or neutral sentiments. The community's response was conveyed through tweets on the Twitter application. Based on the public's response to the policy, sentiment classification can be done using data mining classification techniques. Some research has been carried out on classification techniques using deep learning and machine learning methods. In general, deep learning methods get better results, and this research will be approached using the CNN method. The system stages start from crawling data, labeling, and preprocessing, which consists of cleaning, case folding, tokenization, normalization, removing stopwords and stemming, classification using CNN, and evaluation using 10-Cross Validation. The dataset used is 17.270. The results show that the developed classification system is relatively high, with the highest accuracy of 87%, 93% recall, 93% precision, and 90% F1 score. An in-depth analysis of the classification results and an understanding of sentiment toward rising fuel prices can also provide valuable insights.