<![CDATA[Sentiment classification is essential for analyzing public opinion, particularly on social media issues. One of the main challenges in sentiment classification is the limited amount of training data, which often affects the model's ability to make accurate predictions. This study examines Kaesang Pengarep's appointment as PSI chairman using feature extraction methods such as FastText, TF-IDF, and IndoBERT, alongside the K-Nearest Neighbor (KNN) algorithm. Optimization steps include adding external data, refining text preprocessing, applying data scaling, and tuning parameters. The baseline model achieved 44% accuracy and 39% F1-score using FastText. After optimization and switching to IndoBERT, the optimal model achieved 57% accuracy and 49% F1-score, showing a 10% improvement. These findings demonstrate that optimizations, such as advanced feature extraction and parameter tuning, significantly impact sentiment classification. Future research could focus on advanced optimization techniques to address data limitations and enhance sentiment analysis performance. Keywords: Sentiment Classification, Model Optimisation, K-Nearest Neighbor, FastText, TF-IDF, IndoBERT.]]>
Copyrights © 2025
