Gated Recurrent Unit (GRU) for Sentiment Classification on Imbalanced Data: The COVID-19 Vaccine Program in Twitter

Abstract— The initial implementation of the COVID-19 vaccination by the Indonesian government sparked mixed reactions from the public, ranging from strong support to fierce opposition. These differing opinions influenced individuals' decisions to either accept or refuse the vaccination program for themselves or their families. Public sentiment, expressed through posts, comments, or status updates, provides valuable insights into vaccine acceptance or rejection. This study conducts sentiment analysis using deep learning techniques, specifically employing the Gated Recurrent Unit (GRU) method on Twitter data. The dataset consists of three sentiment classes: positive, negative, and neutral. The Word2Vec word embedding model was used as input and trained on a COVID-19 vaccination sentiment dataset collected from Twitter. Since the classes in the existing data tweets are imbalanced, some other steps are required to improve the classification. The best-performing model achieved an F1-score of 66% and an accuracy of 69%. This classification model effectively addresses the class imbalance problem, delivering competitive results compared to other methods.