Arion data processing

Abstract

This project was carried out in partnership with Arion Bank focuses on using data science methods to analyze customer feedback in text form for assessing customer satisfaction. The main objectives involve gathering data preparing it developing models, categorizing sentiments, evaluating models and creating a prototype for integration into Arion Banks operations.The team. Processed a wide range of text data from Bland.is and Arion Bank to ensure a diverse and thorough dataset. They trained various models like Naive Bayes, Support Vector Machine (SVM) and IceBERT using Icelandic text data. The performance of each model was assessed using metrics such as precision, recall and F1 score.The results showed that while BERT achieved the performance with an average F1 score of 94.67% the SVM model also performed well with an average F1 score of 90.63%. These outcomes exceeded our expectations proving the accuracy and readiness of the models for practical use by Arion Bank in assessing customer sentiment effectively.Although the results are promising there are limitations to the models such as misclassification of sentiments and a relatively small dataset size. To enhance performance expanding the dataset and refining the models to better handle complex or ambiguous sentiments would be beneficial.This initiative showcases how sophisticated data analysis methods can boost customer happiness by offering perspectives on customer opinions. Effectively incorporating these strategies could greatly enhance decision making and customer engagements at Arion Bank.