Sentiment Analysis of YouTube Comments on the Internet Data Quota Rollover Policy Using Support Vector Machine and Naïve Bayes Algorithms

Abstract

The growing use of the internet has made internet data quota an important need for supporting communication, education, work, business, and entertainment. The issue of unused quota expiring at the end of its validity period has generated various public responses, especially following the introduction of an internet data quota rollover policy as a form of protection for user rights. This study aims to analyze public sentiment toward the internet data quota rollover policy and compare the performance of Support Vector Machine (SVM) and Naïve Bayes algorithms in sentiment classification. The research data consisted of 3,281 public comments collected through scraping from a Liputan6 YouTube source discussing the internet data quota rollover policy. The data underwent preprocessing, including case folding, cleaning, tokenizing, normalization, stopword removal, and stemming. The text was then converted into numerical representations using Term Frequency–Inverse Document Frequency (TF-IDF) and classified into three categories: positive, negative, and neutral. Model evaluation used accuracy, precision, recall, F1-score, and a confusion matrix. The results show that neutral sentiment dominated at 64.54%, followed by negative sentiment at 25.51% and positive sentiment at 9.95%. SVM achieved an accuracy of 91%, while Naïve Bayes achieved 74%. SVM also performed better in most sentiment classes, with F1-scores of 0.94 for neutral, 0.91 for negative, and 0.74 for positive sentiment. Meanwhile, Naïve Bayes produced F1-scores of 0.83, 0.52, and 0.09, respectively. Thus, SVM performed better than Naïve Bayes in classifying public comments about the internet data quota rollover policy.