PERANCANGAN SISTEM EKSTRAKSI DAN ANALISIS SENTIMEN KOMENTAR PENGGUNA MYPERTAMINA DI TWITTER

• Main Author: W. V. Hutabarat, Andita
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/72680
• Institution: Institut Teknologi Bandung

To ensure that the distribution process of subsidized fuel is more well-targeted, PT Pertamina has developed an application called MyPertamina and implemented a policy that requires customers to register their vehicles at MyPertamina to buy subsidized fuel. The increasing number of MyPertamina users has led to an increasing number of reviews related to the use of MyPertamina. Reviews of MyPertamina fill various social media channels, including Twitter. However, until now, the analysis of user perceptions through social media has not been optimal. Therefore, a better user sentiment mapping is needed. This study was conducted to answer this need by building a text mining model and designing an information system prototype that can extract and analyze sentiments from tweets related to MyPertamina. This research adopts the CRISP-DM methodology, which consists of business understanding, data understanding, data preparation, modeling, evaluation, and deployment. The data obtained for model development reached 6,920 tweet data. The data went through the stage of data understanding and data preparation. Each data was classified into one of three sentiment categories, namely positive, negative, and neutral. After data preparation, 2,057 data were used for model development. The models tested in this study consist of Support Vector Machine (SVM), Multinomial Naïve Bayes, Gaussian Naïve Bayes, Long Short-Term Memory (LSTM), and Bidirectional Long Short-Term Memory (Bi-LSTM) algorithms. The model that produced the best evaluation score and was selected for prototype development is the SVM model with an accuracy score of 83.74%, weighted precision of 83.96%, weighted recall of 83.74%, and weighted F1-score of 83.72%. The prototype is used for extracting and predicting sentiment for new datasets, which can then be visualized in the form of graphs, word clouds, and tables according to the needs and can be saved on system user’s device.