Abstract: This investigation assesses the customer experience (CX) associated with the Société Générale Algeria (SGA) mobile banking application by leveraging sentiment analysis. Utilizing both machine learning (ML) and deep learning (DL) methodologies, we examined sentiments articulated by users within reviews on application stores and responses gathered from satisfaction surveys. Our study contrasts the efficacy of conventional ML models, specifically Support Vector Machines (SVM) and Naive Bayes, with more advanced DL architectures, namely Long Short-Term Memory (LSTM) networks and Transformer-based models like BERT. The analysis encompassed a multilingual dataset comprising 15,000 comments, mirroring the linguistic heterogeneity of SGA’s clientele, including inputs in Arabic, French, English, and Algerian Arabic dialect. Findings suggest that DL models, particularly BERT (employed here synergistically with SVM), generally exhibit superior performance over traditional models concerning accuracy and the capacity to discern emotional subtleties. However, the overall performance metrics (e.g., a macro F1-score of 0.39 for the BERT+SVM configuration) point towards inherent challenges potentially stemming from task complexity or specific data characteristics. These outcomes underscore the utility of sophisticated natural language processing techniques for acquiring a granular understanding of user sentiments within intricate multilingual environments.

Keywords: Sentiment Analysis, Deep Learning, Natural Language Processing, BERT, LSTM, Mobile Banking Application, Financial Technology

Pages: 23-27

WSEAS Transactions on