Klasifikasi Sentimen Ulasan Pengguna Aplikasi Alfagift Menggunakan Metode Support Vector Machine Berbasis TF-IDF
TL;DR
This study aims to classify the sentiment of Alfagift user reviews into three classes: Positive, Negative, and Neutral, using the Support Vector Machine (SVM) method with Term Frequency-Inverse Document Frequency (TF-IDF) feature representation.
Abstract
The rapid development of digital technology has driven the increasing use of mobile applications in the retail sector, one of which is the Alfagift application developed by PT Sumber Alfaria Trijaya Tbk. User reviews on Google Play Store contain valuable information regarding user satisfaction and complaints, yet their large volume makes manual analysis inefficient, necessitating an automated approach. This study aims to classify the sentiment of Alfagift user reviews into three classes: Positive, Negative, and Neutral, using the Support Vector Machine (SVM) method with Term Frequency-Inverse Document Frequency (TF-IDF) feature representation. A total of 4,937 reviews were collected through web scraping and processed through preprocessing stages including case folding, slang normalization, tokenization, stopword removal, and stemming. Feature extraction used TF-IDF with unigram-bigram configuration and a maximum of 10,000 features, with an 80:20 data split ratio. The model was evaluated using a confusion matrix and 10-fold Stratified Cross Validation. The results show a test accuracy of 87.15% and a cross-validation average of 87.28% (standard deviation 1.0368%). The Positive and Negative classes were classified well (F1-score 0.9172 and 0.8676), while the Neutral class was poorly classified due to extreme class imbalance, representing only 4.7% of the entire dataset.