<![CDATA[This study explores user reviews of the Wondr mobile banking application to identify factors that influence user experience and service quality. The dataset, obtained from the Google Play Store, was processed through several preprocessing steps, including normalization, stopword removal, and stemming. Two topic modelling methods were applied: Latent Dirichlet Allocation (LDA) as a probabilistic baseline and BERTopic as an embedding-based approach. The LDA model was evaluated using coherence scores to determine the most suitable number of topics, while BERTopic was assessed based on topic distribution, interpretability, and additional coherence analysis. The results show that BERTopic produces more semantically meaningful and contextually rich topics, particularly in capturing short-text user reviews. Although BERTopic achieves lower overall coherence compared to LDA, certain topics demonstrate high semantic consistency, especially for well-defined issues such as login verification problems. The analysis reveals that most user feedback is concentrated on positive user experience, while critical issues related to login verification and system errors remain significant concerns. These findings provide actionable insights for improving mobile banking services and demonstrate the effectiveness of embedding-based topic modeling in financial text analytics. These findings highlight a trade-off between statistical consistency and semantic richness in topic modeling approaches. The results provide actionable insights for improving mobile banking services and demonstrate the effectiveness of combining probabilistic and embedding-based methods in financial text analytics.]]>
