ANALISIS SENTIMEN ULASAN PENGGUNA APLIKASI IDENTITAS KEPENDUDUKAN DIGITAL MENGGUNAKAN SUPPORT VECTOR MACHINE RANDOM FOREST
TL;DR
The sentiment of IKD user reviews is analyzed and the effectiveness of Support Vector Machine (SVM) and Random Forest is compared to provide a data-driven basis for the Directorate General of Population and Civil Registration to prioritize feature improvements and enhance IKD service quality.
Abstract
Abstract: The Digital Population Identity (IKD) application allows citizens to access population documents electronically. Service evaluations can be found in user reviews on the Google Play Store, but manual analysis is challenging due to their large and unstructured volume. This study analyzes the sentiment of IKD user reviews and compares the effectiveness of Support Vector Machine (SVM) and Random Forest. Reviews were collected through web scraping, labeled positive or negative by specialists, and preprocessed (cleaning, case folding, tokenizing, normalization, stopword removal, stemming) before being converted into numerical features using TF-IDF weighting. The weighted features were grouped using K-Means clustering to map sentiment tendencies prior to classification. SVM classified sentiment by identifying a maximum-margin hyperplane across classes, while Random Forest combined the predictions of numerous decision trees through majority voting. Both models were trained and tested with a 70:30 stratified split and evaluated using a confusion matrix. From 9,243 scraped reviews collected between June 2025 and June 2026, 5,714 remained after cleaning. SVM achieved 93.64% accuracy with weighted precision, recall, and F1-score of 0.94, while Random Forest achieved 93.41% accuracy with an F1-score of 0.93. Sentiment was dominated by the negative class at 75.5%, mainly complaints about application performance. These results provide a data-driven basis for the Directorate General of Population and Civil Registration to prioritize feature improvements and enhance IKD service quality. Keywords: sentiment analysis; IKD; SVM; Random Forest Abstrak: Aplikasi Identitas Kependudukan Digital (IKD) memungkinkan masyarakat mengakses dokumen kependudukan secara elektronik. Ulasan pengguna di Google Play Store memuat penilaian layanan, namun jumlahnya besar dan tidak terstruktur sehingga sulit dianalisis secara manual. Penelitian ini menganalisis sentimen ulasan pengguna IKD dan membandingkan kinerja Support Vector Machine (SVM) dengan Random Forest. Data ulasan dikumpulkan melalui web scraping, diberi label positif dan negatif oleh pakar, lalu dipraproses (cleaning, case folding, tokenizing, normalisasi, stopword removal, stemming) sebelum diubah menjadi fitur numerik dengan pembobotan TF-IDF. Fitur hasil pembobotan dikelompokkan dengan K-Means Clustering untuk memetakan kecenderungan sentimen sebelum klasifikasi. SVM mengklasifikasikan sentimen dengan mencari hyperplane bermargin maksimum antar kelas, sedangkan Random Forest menggabungkan prediksi banyak decision tree melalui majority voting. Kedua model dilatih dan diuji dengan pembagian data 70:30 stratified serta dievaluasi menggunakan confusion matrix. Dari 9.243 ulasan hasil scraping pada rentang Juni 2025 sampai Juni 2026, tersisa 5.714 ulasan setelah pembersihan. SVM memperoleh akurasi 93,64% dengan precision, recall, dan F1-score tertimbang 0,94, sedangkan Random Forest memperoleh akurasi 93,41% dengan F1-score 0,93. Sentimen didominasi kelas negatif sebesar 75,5%, terutama keluhan pada kinerja aplikasi. Hasil penelitian ini menjadi dasar evaluasi berbasis data bagi Direktorat Jenderal Kependudukan dan Pencatatan Sipil untuk memprioritaskan perbaikan fitur dan meningkatkan kualitas layanan IKD. Kata kunci: analisis sentimen; IKD; SVM; Random Forest