Sentiment Analysis for Super Applications in Indonesia: A Case Study of Gov2Go App

Recently, the Government of the Republic of Indonesia assessed that the number of applications (approximately 24,000) spread across various ministries and regional institutions were less effective and efficient. Therefore, the government will prepare a super application to replace thousands of applications to support electronic government and will use 2,700 data centers that can produce data integration according to the one-data policy concept. To support this initiative, our research proposes a sentiment analysis based on Expectation-Confirmation Theory (ECT). The proposed framework consists of several stages, namely, data preprocessing (cleaning of input data), tokenization, clustering (using a combination of K-Means and Elbow methods to determine the best K value), TF-IDF feature extraction, and classification based on two classification models for comparison, namely Long Short-Term Memory (LSTM) and Naïve Bayes. The ECT constructs used are expectation, confirmation, and satisfaction. We used comment data (review) from the Gov2Go application on the Google Play Store as a case study. The experimental results show that the LSTM model achieves an accuracy value of 97.49%, 3% better than the Naïve Bayes method. This indicates that the proposed framework can effectively detect the polarity of user sentiment and thus can be used as a recommendation to improve application features.