Integration of BERT and SVM in Sentiment Analysis of Twitter/X Regarding Constitutional Court Decision No. 60/PUU-XXII/2024

This research analyzes public sentiment towards the Indonesian Constitutional Court's decision No. 60/PUU- XXII/2024 by utilizing natural language processing techniques using the BERT (Bidirectional Encoder Representations from Transformers) model and the Support Vector Machine model (SVM). The research methodology includes four stages: data preprocessing, data labeling using BERT, embedding extraction, and SVM model training. The data is taken from the Twitter platform, where various public opinions are reflected in three sentiment categories: positive, neutral, and negative. The preprocessing process results in the removal of approximately 23% of duplicate data, and sentiment labeling shows a dominance of the positive category. Evaluation results from the SVM model training demonstrated varying performance: negative sentiment achieved a Precision of 0.57, Recall of 0.36, and F1-score of 0.44; neutral sentiment had a Precision of 0.81, Recall of 0.62, and F1-score of 0.70; while positive sentiment recorded a Precision of 0.98, Recall of 1.00, and F1-score of 0.99. The model's overall accuracy reached 0.97. These findings indicate that the integration of BERT and SVM is effective for sentiment classification, but improvements are needed in the negative and neutral categories to achieve more balanced performance.