| Abstrak/Abstract |
Recently, Indonesian sentiment analysis is a hot research field, especially for twitter data. Many approaches can be used to classify people’s sentiment on Twitter. One of the best methods in sentiment analysis is machine learning. SVM is a machine learning algorithm that has good performance. However, some of the findings of SVM research have different results, which are no better than other algorithms. It is because each study uses a different SVM configuration. Kernel functions on SVM play an essential role in improving SVM performance. Choosing the sui table Kernel function is also crucial for every particular application of SVM. Therefore, to address the problem, we compared four kernel functions on SVM with TF-IDF, such as Polynomial, Sigmoid, Linear, and Radial Basis Function (RBF) kernel. TF-IDF is employed as a feature extraction and selection to improve SVM performance. We used 4000 tweets about omnibus law issues in Indonesia, and 10-fold cross-validation and confusion matrix were employed to validate and evaluate our models. Based on the experimental results, the RBF kernel function on SVM+TF-IDF using 2000 features achieved the best performance of all feature tests in accuracy, precision, recall, and f-measure with value of 96.61%, 96.70%, 96.58%, 96.60%, respectively. Besides, the RBF kernel function obtained high performance when using low dimensionality of the feature. Meanwhile, the Linear kernel function reached the best performance when using high dimensionality of the feature. |
