Analisis Sentimen Komentar Netizen Terhadap Isu Ijazah Presiden Joko Widodo Menggunakan Naive Bayes Dengan Pelabelan Fuzzy Logic Berbasis Leksikon

Abstract

The controversy surrounding the legitimacy of President Joko Widodo's diploma has sparked widespread discussion on social media, generating diverse public comments with positive, negative, and neutral sentiments. This study aims to analyze Indonesian-language sentiment on the issue using a sequential approach that combines Fuzzy Logic-based labeling with Naive Bayes classification. The methodology encompasses several stages: comprehensive text preprocessing (case folding, tokenizing, filtering, and stemming), term weighting with TF-IDF (Term Frequency-Inverse Document Frequency), automated sentiment labeling using lexicon-based Fuzzy Logic with a conservative threshold of ±2, and supervised classification using the Naive Bayes algorithm. A total of 10,027 comments were collected from three major social media platforms Twitter (X), YouTube, and TikTok spanning the period from December 2024 to May 2025. The dataset was divided into 80% training data (8,021 comments) and 20% test data (2,006 comments). The Fuzzy Logic labeling process, utilizing 28 positive keywords and 36 negative keywords, identified a sentiment distribution of 72.38% neutral, 22.98% positive, and 4.64% negative comments. The Naive Bayes model achieved an overall accuracy of 80.76%, demonstrating excellent performance in detecting neutral sentiment (precision 0.82, recall 0.98) but exhibited lower performance for minority classes: positive sentiment (precision 0.70, recall 0.41) and negative sentiment (precision 0.80, recall 0.04). The class imbalance significantly influenced model predictions, with 85.48% of predictions classified as neutral.