Optimasi Algoritma Support Vector Machine (SVM) Dengan Menggunakan Feature Selection Gain Ratio Untuk Analisis Sentimen

Abstract

The ease of internet access has had a positive impact on the increase in the number of social media users in Indonesia. One of the most widely used applications is X or Twitter. Users often upload posts that contain opinions or sentiments, which trigger debates and discussions. This is interesting to analyze as a study of sentiments or opinions that are trending in society. For this analysis, algorithms such as Support Vector Machine (SVM) are required, which are often used for sentiment analysis. However, SVM lacks in accuracy due to the large number of similar words in the dataset. Words related to sentiment analysis usually have large dimensions, so feature selection is needed to improve SVM performance. This research aims to optimize SVM accuracy by using Feature Selection Gain Ratio. The object of research is a dataset related to the 2017 DKI elections from GitHub. The results showed an increase in SVM accuracy with Feature Selection Gain Ratio. With threshold weight gain ratio > 0.0001 (1732 features), accuracy increases from 61.63% to 71.51%. For threshold weights > 0.002 (518 features), the accuracy increased from 61.63% to 62.79%. Feature selection with Feature Selection Gain Ratio gain ratio produces better accuracy than gain ratio, namely 56.40% with gain ratio and 71.51% with gain ratio for weights > 0.0001. The implications of these findings show that the use of Feature Selection Gain Ratio can improve the accuracy of SVM in sentiment analysis. Social media practitioners can utilize this technique to gain more accurate insights from user data. Further research can focus on developing sentiment analysis algorithms with more sophisticated feature selection techniques for various applications on social media platforms.