Optimization of Deep Learning with FastText for Sentiment Analysis of the SIREKAP 2024 Application

Abstract

This study analyzes public sentiment towards the SIREKAP 2024 application using deep learning. Data was collected from Google Playstore reviews and processed through cleaning, tokenization, and stemming. Word embedding was performed using FastText to capture more accurate word representations, including OOV words. The deep learning models compared were CNN, BiLSTM, and BiGRU. Performance evaluation used accuracy, precision, recall, and F1-score metrics. The results showed that the CNN model with FastText Gensim embedding achieved the highest accuracy of 95.98%, outperforming BiLSTM and BiGRU. This model was more effective in classifying positive and negative sentiments. This study provides insights for developers to improve the performance and public trust in SIREKAP 2024 and opens opportunities for further research with more complex embedding approaches and deep learning models.