<![CDATA[Abstract: User reviews of the Wondr By BNI application on the Google Play Store contain sentiment information that can serve as a basis for evaluating the quality of digital banking services. This study aims to analyze and compare the performance of the Support Vector Machine (SVM) model and Indonesian Bidirectional Encoder Representations from Transformers (IndoBERT) in classifying user review sentiments into three categories: positive, neutral, and negative. Data was collected through web scraping techniques from the Google Play Store during the period of January–December 2025, then processed through preprocessing stages and labeled using the Lexicon-Based method with the InSet dictionary. The SVM model uses TF-IDF feature extraction with variations in the C parameter, while IndoBERT is optimized through a fine-tuning process. Evaluation is carried out using accuracy, precision, recall, and F1-score metrics. Test results show that the SVM model with an optimal parameter of C = 10 achieves an accuracy of 83.29%, whereas IndoBERT reaches an accuracy of 92.70%. IndoBERT outperforms SVM across all evaluation metrics with an average difference of around 9–10%, indicating IndoBERT's ability to understand the context of the Indonesian language more deeply. In conclusion, IndoBERT is a more effective model and is recommended for sentiment classification of Indonesian language digital banking application reviews. Keywords: Sentiment Analysis, Support Vector Machine, IndoBERT, BERT, Wondr By BNI, Text Classification Abstrak: Ulasan pengguna aplikasi Wondr By BNI di Google Play Store mengandung informasi sentimen yang dapat menjadi bahan evaluasi kualitas layanan perbankan digital. Penelitian ini bertujuan menganalisis dan membandingkan kinerja model Support Vector Machine (SVM) dan Indonesian Bidirectional Encoder Representations from Transformers (IndoBERT) dalam mengklasifikasikan sentimen ulasan pengguna ke dalam tiga kategori: positif, netral, dan negatif. Data dikumpulkan melalui teknik web scraping dari Google Play Store pada periode Januari–Desember 2025, kemudian diproses melalui tahapan preprocessing dan dilabeli menggunakan metode Lexicon-Based dengan kamus InSet. Model SVM menggunakan ekstraksi fitur TF-IDF dengan variasi nilai parameter C, sedangkan IndoBERT dioptimalkan melalui proses fine-tuning. Evaluasi dilakukan menggunakan metrik accuracy, precision, recall, dan F1-score. Hasil pengujian menunjukkan bahwa model SVM dengan parameter optimal C = 10 memperoleh akurasi sebesar 83,29%, sedangkan IndoBERT mencapai akurasi sebesar 92,70%. IndoBERT mengungguli SVM pada seluruh metrik evaluasi dengan selisih rata-rata sekitar 9–10%, yang menunjukkan kemampuan IndoBERT dalam memahami konteks bahasa Indonesia secara lebih mendalam. Kesimpulannya, IndoBERT merupakan model yang lebih efektif dan direkomendasikan untuk klasifikasi sentimen ulasan aplikasi perbankan digital berbahasa Indonesia. Kata Kunci: Analisis Sentimen, Support Vector Machine, IndoBERT, BERT, Wondr By BNI, Klasifikasi Teks.]]>
Copyrights © 2026
