| The Indonesian National Police plays an important role in maintaining security and providing services to the public. However, there is still public doubt about the quality of its services. This study aims to analyze public sentiment towards police services using Twitter data with a Natural Language Processing (NLP) approach. A total of 14,718 tweets were collected, and after preprocessing, 13,941 tweets were produced that were worthy of analysis. The data was automatically labeled using the Indonesian lexicon method, resulting in 3,737 positive tweets and 6,869 negative tweets. Text representation was carried out using the Term Frequency–Inverse Document Frequency (TF-IDF) method, then classified with the Support Vector Machine (SVM) algorithm using linear, RBF, and polynomial kernels. The Grid Search results showed that the RBF kernel with parameters C=1000 and gamma=0.1 gave the best performance with an accuracy, precision, and recall of 91.36%. Model evaluation on training and test data ratios (70:30, 80:20, and 90:10) showed the highest accuracy of 91.83% at the 90:10 ratio. 10-fold cross-validation produced an average accuracy of 92.31%, precision of 92.29%, and recall of 92.31%. These results indicate that SVM with RBF kernel is effective in classifying text-based sentiment in Indonesian. |
