查看文章

Phenomenon of online transportation with some problems like crime and fraud in Indonesia triggers pros and cons to Twitter users. This study aims to find out sentiments of the society on online transportation and compare the accuracy of SVM and SVM-PSO with default parameters value. The proposed solution divided the dataset into training and testing data, because some researches only used one dataset that had already been classified. The research data is tweet data, which is obtained through scraping method using Octoparse. A total of 1,852 tweets from 1/1/2019 to 15/10/2019 were divided into 1,130 tweet testing data and 722 tweet training data. Then, RapidMiner was used for analysis process. Analysis positive sentiment using SVM is 62% and negative sentiment is 38%, while in SVM-PSO, positive opinion is 53% and negative opinion is 47%. The results of research using 10 k-fold CV produce accuracy on SVM is 95.46% and AUC is 0.979 (excellent classification), while in SVM-PSO accuracy is 96.04% and AUC is 0.993 (excellent classification). The results show that use of training and testing data on this study can be done and prove that SVM-PSO is better than ordinary SVM, although the parameters value is default.