Analisis Sentimen Pengguna Aplikasi Mobile JKN menggunakan Algoritma Support Vector Machine dan Naive Bayes

Abstract

The Mobile JKN application serves as the primary digital gateway developed by BPJS Kesehatan, enabling National Health Insurance participants to manage administrative needs and access health-related information remotely. Despite its widespread adoption, persistent user grievances—documented through Google Play Store reviews—signal opportunities for service refinement. This research harvested 20,000 user reviews via automated scraping (August 1–December 15, 2025) and retained 18,729 valid entries following a five-stage preprocessing pipeline encompassing cleaning, normalization, tokenization, stemming, and stopword elimination. Feature representation relied on TF-IDF vectorization, while training-set class imbalance was counteracted through SMOTE oversampling. A head-to-head evaluation pitted LinearSVC against Complement Naive Bayes across three sentiment polarities. LinearSVC emerged as the stronger classifier, registering 81.71% accuracy alongside a weighted F1-score of 0.84—surpassing its probabilistic counterpart at 79.85% accuracy and 0.83 weighted F1-score. Both architectures demonstrated robust positive-class detection (F1=0.92) yet faltered on neutral reviews, where overlapping lexical cues between praise and complaint eroded discriminative power. Wordcloud mapping further exposed recurring dissatisfaction markers ("susah", "sulit", "ribet") juxtaposed with appreciation signals ("bantu", "mudah", "bagus"), offering actionable intelligence for BPJS Kesehatan to target specific service pain points.