Sentiment Analysis of Food Poisoning Incidents in Indonesia's Free Nutritious Meal Program Using TF-IDF and SVM

Abstract

Food poisoning incidents associated with the Free Nutritious Meal Program (MBG) have triggered significant public reactions on the social media platform X/Twitter. This study identified and classified public sentiment regarding food poisoning occurrences in the MBG program by applying a Support Vector Machine (SVM) classification framework. Data were crawled from X/Twitter via Tweet Harvest using keywords related to MBG food poisoning incidents between January 2025 and June 2026. A total of 6,929 posts were collected, and 6,902 Indonesian-language records were retained after duplicate removal. The analytical workflow comprised data selection, automated sentiment labeling with IndoBERTweet, text preprocessing, Term Frequency-Inverse Document Frequency (TF-IDF) feature weighting, and classification using the SVM algorithm. Model performance was evaluated through a confusion matrix using accuracy, precision, recall, and F1-score metrics. Based on an evaluation of 1,989 testing records, the SVM model achieved an overall accuracy of 76.42%, exhibiting its strongest performance in the negative sentiment class. However, positive sentiment recorded a low recall of 26.03%, reflecting the difficulty of detecting the minority class. Sentiment distribution analysis revealed that negative sentiment dominated public discourse at 63.78%, followed by neutral sentiment at 29.05% and positive sentiment at 7.17%. These findings indicate that public perceptions of food poisoning incidents in the MBG program were overwhelmingly negative, underscoring critical concerns regarding food safety and program oversight. Beyond contributing to machine learning-based social media analytics, this study provides actionable insights for policymakers to evaluate service delivery and reinforce food safety standards in the MBG program.