SUMMARY
The rise of hate speech on social media can harm various parties, including the candidate for regional head of Karawang Regency in 2020, but because of the large number of comments, the sanctions given to violators are not evenly distributed. To make it easier for Bawaslu to give sanctions to violators and to provide a deterrent effect to the Karawang community so that hate speech does not occur again. Therefore, this study was conducted by classifying positive and negative comments. The methodology used is Knowledge Discovery in Database (KDD) by dividing the data into 4 scenarios. The results obtained state that the Support Vector Machine (SVM) Algorithm with scenario "2" on a linear kernel gets the highest accuracy value of "72.66%". Then the results of the 4 scenarios were optimized by Particle Swarm Optimization which got the highest accuracy value, namely the linear and polynomial kernels in the 4th scenario with 90:10 data sharing of "78.00%". Other evaluation values ??also experienced the same increase, starting from precision, recall, and f1-score. It can be concluded that the Support Vector Machine algorithm optimized with Particle Swarm Optimization can increase the accuracy value.