ANALISIS SENTIMEN ULASAN APLIKASI SMILE INDONESIA MENGGUNAKAN METODE NAIVE BAYES DAN SUPPORT VECTOR MACHINE (SVM)

SENTIMENT ANALYSIS OF SMILE INDONESIA APPLICATION REVIEWS USING NAIVE BAYES AND SUPPORT VECTOR MACHINE (SVM) METHODS

Abstract

Tujuan studi ini adalah untuk mengevaluasi bagaimana pengguna memandang aplikasi digital SMILE Indonesia, sebuah platform layanan publik yang memantau penyampaian layanan kesehatan secara real-time. Menggunakan teknik web scraping, 383 ulasan pengguna dikumpulkan dari Google Play Store dan secara otomatis diklasifikasikan berdasarkan skor penilaian: ulasan dengan skor 1-2 dikategorikan sebagai negatif, ulasan dengan skor 4-5 sebagai positif, dan ulasan dengan skor 3 atau lebih rendah dikecualikan karena kemungkinan ambiguitas. Langkah-langkah pre-processing seperti case folding, pembersihan teks, tokenisasi, penghapusan kata, stemming, dan normalisasi diterapkan pada data yang telah dilabeli. Metode TF-IDF (Term Frequency–Inverse Document Frequency) kemudian digunakan untuk mewakili data secara numerik. Dua algoritma digunakan untuk klasifikasi: Naïve Bayes dan Support Vector Machine (SVM). Hasil evaluasi menunjukkan bahwa SVM mencapai 75% pada keempat metrik, sementara Naïve Bayes mencapai akurasi 79%, presisi 81%, recall 79%, dan F1-score 79%. Uji McNemar menunjukkan bahwa perbedaan kinerja antara kedua model tidak signifikan secara statistik (p > 0.05), meskipun Naïve Bayes memperoleh skor yang lebih tinggi. Penelitian sentimen ini memberikan wawasan tentang bagaimana masyarakat umum memandang layanan publik digital; sementara sikap negatif menekankan kesulitan teknis, sikap positif menyoroti aksesibilitas dan keuntungan praktis. Hasil ini dapat digunakan secara strategis oleh pengembang dan pembuat kebijakan untuk meningkatkan kualitas layanan digital berbasis e-government, terutama di bidang logistik kesehatan.

The purpose of this study is to evaluate how users perceive the SMILE Indonesia digital application, a public service platform that monitors the delivery of health services in real time. Using web scraping techniques, 383 user reviews were collected from the Google Play Store and automatically classified based on rating scores: reviews with scores of 1-2 were categorized as negative, reviews with scores of 4-5 as positive, and reviews with scores of 3 or lower were excluded due to potential ambiguity. Pre-processing steps such as case folding, text cleaning, tokenization, word removal, stemming, and normalization were applied to the labeled data. The TF-IDF (Term Frequency–Inverse Document Frequency) method was then used to represent the data numerically. Two algorithms were used for classification: Naïve Bayes and Support Vector Machine (SVM). Evaluation results show that SVM achieved 75% on all four metrics, while Naïve Bayes achieved 79% accuracy, 81% precision, 79% recall, and 79% F1-score. The McNemar test indicates that the performance difference between the two models is not statistically significant (p > 0.05), although Naïve Bayes achieved higher scores. This sentiment analysis provides insights into how the general public perceives digital public services; while negative attitudes emphasize technical difficulties, positive attitudes highlight accessibility and practical benefits. These results can be strategically utilized by developers and policymakers to improve the quality of e-government-based digital services, particularly in the field of health logistics.