ANALISIS SENTIMEN PADA PENGGUNA APLIKASI DANA MENGGUNAKAN METODE LSTM DAN BERT UNTUK MENINGKATKAN PENGGUNA APLIKASI DANA

Abstract

Sentiment analysis is an important computational technique used to identify opinions and perceptions expressed by users through textual reviews on digital platforms. This research aims to conduct sentiment analysis on user reviews of the DANA digital wallet application collected from the Google Play Store. The dataset was systematically obtained through a web scraping process, resulting in thousands of review entries as the primary data source. The text data underwent several preprocessing stages, including case folding, cleansing, slang normalization, stopword removal, and tokenization, to produce clean and structured text suitable for modeling. The sentiment classification model employed in this study is the LSTM-BERT architecture, in which IndoBERT is utilized to generate contextual word representations based on the Transformer mechanism, while LSTM is used to capture sequential patterns within the textual data. The model was trained using training data and evaluated using validation and testing datasets. Model performance was assessed using accuracy, precision, recall, F1-score metrics, and confusion matrix visualization. The experimental results indicate that the proposed model achieved an accuracy of 0.70, a weighted average F1-score of 0.67, and a macro average F1-score of 0.58. The model performed well in identifying positive and negative sentiment classes, while performance on the neutralclass remained relatively low due to dataset imbalance issues. These findings demonstrate that the LSTM-BERT model is effective for sentiment classification in Indonesian-language review data from fintech applications, although further improvements are required, particularly in addressing class imbalance to enhance overall model performance.