<![CDATA[The development of internet technology and social media has driven the increasing use of sentiment analysis to understand public opinion. This study aims to improve the classification performance of sentiment analysis by proposing a hybrid model that combines FastText-BLSTM and BERT. The dataset used consists of 900 Indonesian-language Netflix app user reviews obtained through crawling using Google Play Scraper. The research stages include text preprocessing, feature extraction using FastText and BERT, and classification using BLSTM, which are then combined in a concatenation layer to produce a richer feature representation. Experimental results show that the FastText-BLSTM-BERT hybrid model provides the best performance with an accuracy of 94.22%, a precision of 95.98%, a recall of 95.68%, and an F1-score of 95.83%. This achievement is superior to the single models of FastText-BLSTM and BERT. The main novelty of this research lies in the integration of contextual embeddings from BERT with subword-level semantic and sequential representations from FastText-BLSTM, which has not been extensively explored in prior studies on Indonesian sentiment analysis. This hybridization demonstrates significant improvement in model generalization and robustness for low-resource language texts]]>
Copyrights © 2026
