Analisis Sentimen Sosial Media X terhadap Isu Pilkada 2024 Menggunakan Word Embedding dan Bidirectional Long-Short Term Memory (Bi-LSTM)

The 2024 regional elections have become a hot political topic on social media platform X (Twitter), with various public opinions ranging from support to criticism of candidates and organizers. To gain a deeper understanding of public sentiment, this study uses a Bidirectional Long Short-Term Memory (Bi-LSTM) model with three types of word representations, namely FastText, GloVe, and Word2Vec. A total of 16,929 tweets were collected using a crawling method with the keyword “2024 Regional Elections” in the period before and after the election. After the preprocessing stage, the data was adjusted using a WordNet-based augmentation method (synonyms and random deletion) to address the imbalance between positive, negative, and neutral sentiment classes. The assessment results show that the three embedding methods demonstrate competitive performance with 96% accuracy. FastText excels in neutral sentiment classification with an F1-score of 0.97, GloVe excels in the negative class with a recall value of 0.99, while Word2Vec shows fairly balanced performance across all classes. This study indicates that the combination of Word Embedding and Bi-LSTM is successful in categorizing public sentiment on political issues on social media platforms, while FastText is better at addressing language variations in Indonesian social media texts.