<![CDATA[Abstrak. Pertumbuhan layanan perbankan digital di Indonesia menjadikan ulasan pengguna pada Google Play Store sebagai sumber penting untuk mengevaluasi kualitas aplikasi, termasuk Bank Jago. Namun, ulasan tersebut bersifat tidak terstruktur, informal, dan mengandung noise sehingga menyulitkan analisis sentimen. Penelitian ini bertujuan memberikan gambaran objektif kecenderungan opini pengguna serta membandingkan kinerja algoritma Support Vector Machine (SVM) dan Neural Network (MLPClassifier). Sebanyak 10.000 ulasan dikumpulkan melalui scraping dan direduksi menjadi 7.946 ulasan setelah penghapusan duplikasi. Data diproses melalui tahapan preprocessing meliputi cleaning, case folding, normalisasi slang, tokenisasi, stopword removal, dan stemming. Pelabelan sentimen dilakukan menggunakan lexicon InSet, sedangkan ekstraksi fitur menggunakan CountVectorizer berbasis Bag-of-Words. Hasil penelitian menunjukkan bahwa SVM memperoleh akurasi tertinggi sebesar 91,2%, lebih unggul dibandingkan Neural Network dengan akurasi 89,8%. Temuan ini menegaskan bahwa pemilihan preprocessing dan representasi fitur yang tepat berperan penting dalam meningkatkan performa analisis sentimen pada ulasan aplikasi perbankan digital. Abstract. The growth of digital banking services in Indonesia has made user reviews on the Google Play Store an important source for evaluating application quality, including Bank Jago. However, these reviews are unstructured, informal, and noisy, creating challenges for sentiment analysis. This study aims to provide an objective overview of user sentiment and to compare the performance of Support Vector Machine (SVM) and Neural Network (MLPClassifier). A total of 10,000 reviews were collected through scraping and reduced to 7,946 reviews after duplicate removal. The data were processed through preprocessing stages including cleaning, case folding, slang normalization, tokenization, stopword removal, and stemming. Sentiment labeling was conducted using the InSet lexicon, while feature extraction employed a Bag-of-Words approach with CountVectorizer. The results show that SVM achieved the highest accuracy of 91.2%, outperforming the Neural Network model with 89.8%. These findings highlight the importance of appropriate preprocessing and feature representation for improving sentiment analysis performance in digital banking application reviews.]]>
