Penerapan Algoritma Naïve Bayes Dalam Analisis sentiment Masyarakat Terhadap STMIK Widya Cipta Dharma

Abstract

This study applies the Naïve Bayes algorithm to analyze public sentiment toward STMIK Widya Cipta Dharma using Google Maps reviews as the primary data source. The research aims to classify community perceptions into three categories: positive, neutral, and negative. The methodology follows the CRISP-DM framework, incorporating stages such as data preprocessing (text cleaning, stopword removal, and stemming), TF-IDF for feature extraction, and SMOTE to address class imbalance. Sentiment labels were derived from a combination of review ratings (1–5 stars) and textual content. Results indicate that Naïve Bayes achieved 91% accuracy in classifying the majority (positive) class but struggled with minority classes (neutral and negative), yielding 0% precision and recall for these categories. After applying SMOTE, recall for the negative class improved to 100%, although overall accuracy dropped to 38%, reflecting a trade-off between balanced class recognition and model performance. The study highlights the algorithm's effectiveness in handling large-scale text data but underscores challenges in managing imbalanced datasets. These findings provide actionable insights for STMIK Widya Cipta Dharma to enhance service quality and institutional image by leveraging public feedback. Future research could explore hybrid algorithms or advanced preprocessing techniques to optimize sentiment analysis accuracy across all classes.