<![CDATA[This study aims to classify Indonesian-language news using the Long Short-Term Memory (LSTM) method and to evaluate its performance through accuracy, precision, recall, and F1-score metrics. The dataset consists of 48,634 news titles collected from various national and regional portals, covering five main categories: finance, travel, health, food, and sports. The research process involves several text preprocessing stages-tokenization, stop-word removal, normalization, and stemming-followed by feature representation using word embedding and the design of the LSTM model architecture. The model's performance is assessed using a confusion matrix along with additional validation through cross-validation to ensure result consistency. The LSTM model demonstrates strong performance, achieving 90% accuracy, 89% precision, 88% recall, and 89% F1-score, indicating its capability to capture semantic patterns and contextual dependencies in textual data effectively. In addition, LSTM outperforms the baseline method with a 6% increase in accuracy, reinforcing its reliability for Indonesian text classification tasks. Overall, the findings confirm that the combination of optimal preprocessing techniques and a well-designed LSTM architecture enhances the performance of the news classification system and offers significant potential for various text analysis applications in the digital information era.]]>
Copyrights © 2026
