<![CDATA[Social media, particularly Twitter, has evolved into a dynamic arena for discussions on religious issues in Indonesia. Interfaith tolerance is one of the topics that most frequently elicits a wide range of responses, from support to hate speech. This study designs a five-class sentiment scheme, consisting of Positive/Neutral, Neutral-Abusive, and Negative tweets divided into three intensity levels (Weak, Moderate, and Strong), and classifies them using the Random Forest algorithm. The dataset used is the Indonesian Abusive and Hate Speech Twitter Text available on the Kaggle platform, consisting of 13,169 tweets with dual labels. Sentiment labels were created based on a combination of the HS and HS_Religion columns and hate speech intensity levels: Weak, Moderate, and Strong. Tweets without hate speech and unrelated to religion are considered positive or neutral, while tweets with HS_Religion=1 are classified as negative and grouped into three intensity levels. Prior to modeling, the text undergoes slang normalization, removal of inappropriate words, Nazief-Adriani stemming, and feature extraction using TF-IDF bigrams. Results from 10-fold cross-validation show an accuracy of 66.0%, macro precision of 52.1%, macro recall of 56.1%, and macro F1-Score of 51.3%, comparable to SVM (F1 52.7%) and Naive Bayes (31.8%), with differences between models assessed statistically using the McNemar test.]]>
Copyrights © 2026
