Twitter had 24 million users in Indonesia at the beginning of 2023. Despite having fewer users than other platforms, its fast and instant nature makes Twitter a significant source of information dissemination. Tweets shared on Twitter offer various advantages. However, it also has negative consequences, including the dissemination of fake news, instances of cyberbullying, and the expression of hate speech. Specifically, hate speech employs offensive language to discriminate against an individual or group based on race, ethnicity, nationality, religion, gender, sexual orientation, or other personal attributes, leading to discord. Such behavior comes under the jurisdiction of various legal statutes, including the Constitution, the Criminal Code, and the ITE Law. The primary objective of this research is to categorize tweets shared on Twitter into hate speech and non-hate speech sentiments, utilizing a Support Vector Machine (SVM) algorithm based on a dataset of 5,000 tweets. This research involved data preprocessing, labeling, feature extraction using TF-IDF, model training (80%), and testing (20%). The final stage includes enhancing SVM parameters through GridSearch and cross-validation methods (GridSearchCV), followed by analysis using a Confusion Matrix with the Matplotlib Library. Radial Basis Function (RBF) kernels, defined by parameters C=10 and gamma=0.1, exhibited the highest performance among SVM models, boasting an 84% accuracy. The RBF kernel also attained 85% precision, 97% recall, and a 91% F1-score for hate speech identification. In conclusion, the evaluation of SVM kernel performance highlights the superiority of RBF kernels in achieving the highest accuracy, complemented by nuanced insights into hate speech precision, recall, and F1-score values across various kernel types.
Riadi et al. (Sun,) studied this question.