<![CDATA[Post-marketing vaccine safety surveillance is one of the fundamental pillars in maintaining public confidence in national and global vaccination programs. The Vaccine Adverse Event Reporting System (VAERS), jointly managed by the U.S. Food and Drug Administration (FDA) and the Centers for Disease Control and Prevention (CDC), receives hundreds of thousands of vaccine adverse event reports each year, most of which are submitted as free-text narratives that are difficult to analyze manually. This study applies the Multinomial Naive Bayes algorithm with a Bag of Words (BoW) representation to classify the severity of vaccine adverse events based on the narrative text of VAERS patient reports from 2025.The dataset consists of 40,702 reports categorized into three severity classes: Mild (34,693 reports, 85.2%), Moderate (4,827 reports, 11.9%), and Severe (1,182 reports, 2.9%). Each report was assigned to a severity category according to the clinical criteria explicitly specified in the structured VAERS data fields. The text preprocessing stage included converting all text to lowercase and removing non-alphanumeric characters. Text features were represented using CountVectorizer with the 5,000 most frequent features and English stop-word removal.Model performance was evaluated using 5-fold Stratified K-Fold Cross-Validation with precision, recall, F1-score, and accuracy as evaluation metrics. The experimental results achieved an average accuracy of 71.39%, precision of 86.70%, recall of 71.39%, and F1-score of 75.95%, with consistent performance across all folds, indicating the model's strong generalization capability on previously unseen data. The findings of this study are expected to provide a foundation for developing resource-efficient and accountable medical report analysis tools, particularly within artificial intelligence-powered vaccine safety surveillance systems]]>
