KOMPARASI MODEL DEEP LEARNING PADA ULASAN APLIKASI DRAMABOX DI GOOGLE PLAY STORE

Abstract

The rapid growth of streaming platforms and the increasing popularity of micro-drama have led to the emergence of applications such as DramaBox, which generate a large number of user reviews as a source of public opinion. However, sentiment analysis on streaming application reviews is still limited, particularly in comparing the performance of deep learning models. This study aims to compare the performance of Bi-LSTM and IndoBERT models in conducting sentiment analysis on user reviews of the DramaBox application. The data were collected through web scraping from the Google Play Store, resulting in a total of 13,701 user reviews. The data then underwent preprocessing before being divided into data train and data test. Model evaluation was carried out using accuracy, precision, recall, F1-score, confusion matrix, and 5-fold cross-validation. The results show that the IndoBERT model achieved better performance with an accuracy of 0.90 compared to 0.87 for Bi-LSTM, and also demonstrated higher stability based on cross-validation results. Therefore, IndoBERT is considered more effective in understanding the context of the Indonesian language and producing more accurate sentiment analysis on the DramaBox user review dataset.