Penilaian Perspektif Pengguna Aplikasi Pinjaman Online dengan Pemodelan Topik Latent Dirichlet Allocation pada Analisis Sentimen Menggunakan Extreme Gradient Boosting

Abstract

Every application user has the right to provide a review regarding their experience when using the application service. These reviews is a source of information for application developers to improve the quality of their services. Therefore, this study performs a combination of sentiment analysis and topic modeling to evaluate user perspectives. At the feature extraction stage, TF-IDF is used to convert words into vectors with appropriate weights. Additionally, LDA topic modeling was used to identify key topics in the data. XGBoost to analyze sentiment based on previously identified topics. This research uses a dataset in the form of 3500 user reviews of the AdaKami application, which was obtained through a data crawling process from the Google Play Store. Evaluation of model quality uses the Confusion Matrix which produces an accuracy of 92% for the application data aspect, 90% for the debt payment aspect, and 93% for the fund disbursement aspect. Next, the NRS method is used to calculate application reputation, with the highest NRS in the fund disbursement aspect at 93.21% and the lowest NRS in the debt payment aspect at -26.71%. Based on the results of the training data, it was found that LDA topic modeling provides suitable topics for sentiment analysis based on aspects using XGBoost. This results in maximum performance and high accuracy in this research.