<![CDATA[The Directorate General of Immigration launched the M-Paspor app for online passport applications. Although it has been downloaded more than 1 million times, its 2.5 rating indicates user dissatisfaction. Review analysis is necessary for developers to understand the issues and improve the app to enhance the overall user experience. Therefore, this study conducts sentiment analysis on M-Paspor app reviews using a combination of Bidirectional Encoder Representations from Transformers (BERT) embedding and LSTM to classify user opinions. The advantage of BERT lies in its ability to understand the deep context of text, while LSTM excels in handling sequential data. LSTM is used as a classification method because it can capture long-term patterns in sequential data through memory management with cell states and three main gates, enabling it to continuously understand sentence context to support sentiment analysis. In this study, labeling consists of three classes: positive, negative, and neutral, using the lexicon method. The LSTM-BERT model shows consistent and higher accuracy values with a smaller proportion of training data. Testing results with a confusion matrix show the highest accuracy of 91.33% on a 70%:30% data split.]]>
Copyrights © 2025
