Abstract
This study investigates customer sentiment and satisfaction within the South African retail
banking sector, emphasising the influence of digital platforms and evolving customer expectations.
Sentiment analysis was conducted on customer reviews from Hellopeter using the Valence
Aware Dictionary for Sentiment Reasoning (VADER) algorithm and various machine learning
models, including multinomial na¨ıve bayes (MNB), logistic regression (LR), support vector machine
(SVM), and ensemble models. The ensemble model achieved the best performance, with
an average F1-score of 90%, accuracy of 91%, precision of 92%, and recall of 87% across all
sentiment classes. Thematic analysis, facilitated by Bidirectional Encoder Representations from
Transformers for Topic Modeling (BERTopic) and Braun & Clarke’s framework, identified 41
positive themes, such as favourable customer experiences, and 76 negative themes related to
account services and general banking issues. Customer satisfaction was evaluated using the Expectation
Disconfirmation Theory (EDT), comparing theme-specific sentiment scores against
industry benchmarks from the 2023 South African Banking Sentiment Index Report. Results
indicated positive disconfirmation in areas like customer service and Coronavirus Disease 2019
(COVID-19) relief efforts, while negative disconfirmation was observed in credit card applications,
branch communication, and loan management. This research contributes to a deeper
understanding of customer experience and satisfaction by integrating EDT with sentiment analysis
and applying thematic analysis to extensive customer data. The findings provide actionable
insights for service enhancements in retail banking, and future studies could explore additional
data sources and incorporate frameworks such as Service Quality (SERVQUAL) model for a
more comprehensive assessment of customer satisfaction in this sector.