Analisis Sentimen Pasien Pada Ulasan Layanan Puskesmas Sekota Pekanbaru Menggunakan Metode Naive Bayes Classifier

Abstract

The rapid development of information technology has encouraged the public to share opinions regarding healthcare services through digital platforms such as Google Maps. These patient reviews can be utilized to evaluate the quality of public health center (Puskesmas) services in a more objective and data-driven manner. This study aims to analyze patient sentiment toward Puskesmas services across Pekanbaru City using the Naïve Bayes Classifier (NBC) method. The research data were collected through a web scraping technique from Google Maps, resulting in 856 patient reviews. The research stages included data preprocessing (case folding, cleansing, normalization, tokenizing, stopword removal, and stemming), sentiment labeling (positive, negative, and neutral), TF-IDF weighting, classification using NBC, and model evaluation. The initial sentiment distribution consisted of 480 neutral reviews, 259 positive reviews, and 117 negative reviews. To address data imbalance, the SMOTE method was applied. The evaluation results using a 70% training and 30% testing split showed an accuracy of 71.60%. After applying SMOTE, the accuracy increased to 78.21%, while the implementation of Chi-Square feature selection produced the highest accuracy of 80.56%. Meanwhile, PCA and LDA achieved accuracies of 71.06% and 57.41%, respectively. Word cloud visualization and TF-IDF analysis revealed dominant words such as “service,” “friendly,” “good,” and “bad.” The findings indicate that the Naïve Bayes method is effective for classifying patient review sentiments, particularly when combined with Chi-Square feature selection. This study is expected to provide a basis for improving the quality of Puskesmas services in Pekanbaru City.