Analyzing Public Opinion Based on Emotion Labeling Using Transformers

Abstract

This research aimed to do sentiment analysis by conducting text classification targeting six basic human emotions (fear, anger, joy, sadness, disgust, and surprise) using state-of-the-art Natural Language Processing (NLP) technique called ‘Transformers'. More than 1000 tweet data are obtained from Twitter on the issue of the mudik prohibition policy issued by the government of Indonesia in May 2021. The result showed that most people are feeling sad (47%) and surprised (24%) about the mudik prohibition policy. The sad feeling is related to the publics' inability to come back to their hometown and missing their families there. Whereas the ‘surprised’ feelings are due to the contradiction of the mudik prohibition policy with other policies such as the opening of tourist attractions and malls. Our result also showed that the model can accurately predict and have high confidence in predicting the emotions even when the texts do not contain obvious words that are strongly associated with certain emotions. The average confidence score on the prediction is pretty high at 0.82 with most of the predictions having a confidence score higher than 0.95.