Analysis of Banking Application Reviews Using a Topic-based Sentiment Analysis Approach with Rule-based Lexicon and LDA

This study analyzed banking user reviews using the Livin' by Mandiri app as a case study to understand user perceptions of digital banking services. A topic-based sentiment analysis approach was implemented by integrating a rule-based lexicon method and Latent Dirichlet Allocation (LDA). Text preparation steps such as cleaning, tokenization, stopword removal, and stemming were applied to 13,539 Google Play Store reviews that were gathered between January and June 2025. Sentiment labeling using the INSET lexicon indicated that 58.7% of reviews were negative, while 41.3% were positive. Topic modeling identified five main themes representing key service aspects, with authentication issues, such as login and facial verification failures, emerging as the dominant topic in negative reviews and achieving the highest coherence score of 0.486909. Model evaluation was conducted using coherence measurement and manual validation to ensure interpretative consistency. The findings indicated that authentication system stability significantly influenced negative user perceptions, whereas transaction efficiency and ease of use contributed to positive evaluations. This approach provided interpretable insights to support data-driven service improvement in digital banking applications.