Fake News Classification Using Sentence-Level Features as a Counterbalance to Prevent Spread on Social Media
Abstrak
The spread of fake news on social media has become a serious issue that can influence public opinion and social stability. This study aims to develop a fake news classification system by focusing on sentence-level linguistic features. The six features used in this research are Parts of Speech (POS) Tagging, average tweet length, punctuation frequency, function words, phrase structure, and sentence complexity. The data were collected from reliable sources such as TurnBackHoax.id, Komdigi, Kompas, and Twitter, covering the period from 2020 to 2025. The research process involved several steps including data preprocessing, document similarity analysis using TF-IDF and cosine similarity, and classification using five machine learning models such as Support Vector Machine (SVM), KNearest Neighbor (KNN), Decision Tree, Naive Bayes, and Multilayer Perceptron (MLP). The models were evaluated using 10-fold cross-validation. Based on the results, the SVM model with an RBF kernel achieved the best performance with an F1-score of 0.844. The system also successfully classified real-time Twitter data, making it useful for detecting new fake news that may not exist in the dataset. This research is expected to contribute to the development of more effective fake news detection systems in the future.