CLUSTER ANALYSIS OF DANA AND GOPAY E-WALLET SERVICE REVIEWS USING TF–IDF AND K-MEANS/K-MEDOIDS: AN ECONOMIC AND MANAGEMENT STRATEGY PERSPECTIVE

Abstract

This study aims to (1) identify the thematic structure of Indonesian-language user reviews on digital wallet services, (2) compare the clustering quality produced by alternative unsupervised methods, and (3) derive actionable operational insights for service management and financial performance. A quantitative text-mining design is employed, using an unsupervised learning framework on Indonesian-language Google Play Store reviews collected over the 2021–2025 period, comprising 553.500 reviews for DANA and 265.632 reviews for GoPay. The corpus is obtained via staged web scraping, cleaned and normalized, then represented with TF–IDF uni- and bigram vectors. Two clustering algorithms, K-Means and K-Medoids with cosine distance, are applied separately to the DANA, GoPay, and combined corpora. Cluster quality is evaluated using internal validity indices, including the Silhouette while thematic interpretation relies on dominant term inspection and supporting visualizations. The findings show that DANA’s main themes revolve around user experience, ease and reliability of transactions, OTP-based security and access barriers, and customer support, whereas GoPay reviews emphasize transactional operations, balance accessibility, and promotions or perceived service levels. The evaluation results indicate that DANA exhibits more heterogeneous thematic segments than GoPay, implying the need for more differentiated targeting and service strategies, while GoPay can rely on a relatively more homogeneous user base.