<![CDATA[The 2024 Jakarta Regional Head Election (Pilkada Jakarta) is a critical issue that requires an in-depth understanding of public sentiment. This platform generates complex, unstructured text with informal language and ambiguity, posing challenges alongside the lack of local context-specific datasets and inaccuracies in traditional sentiment analysis models. Analyzing sentiment for the Pilkada is crucial for evaluating public response to policies, aiding political strategy, and improving governance. Current systems struggle with complex data and class imbalance (dominant neutral sentiment), leading to underrepresented information. This study addresses these issues by constructing a sentiment analysis system using four deep learning models: IndoBERT, LSTM, CNN, and GRU. The procedure encompassed data acquisition from X, preprocessing, model training, and assessment based on accuracy, precision, recall, and F1-score. The CNN model achieved the highest accuracy of 83.37%, followed by LSTM at 82.61%, GRU at 82.30%, and IndoBERT at 80.77%. All models achieved the accuracy benchmark of a minimum of 80%, however the neutral class continues to pose a challenge. Research contributions include a deep learning-based sentiment classification system that can be implemented in local political opinion analysis, as well as recommendations for using hybrid models like IndoBERT + CNN for further research.]]>
Copyrights © 2025
