CONVERSATIONAL BANKING CHATBOT USING NLP AND SUPERVISED MACHINE LEARNING MODEL

Abstract

Chatbots are intelligent systems designed by Artificial Intelligence (AI) and are upgraded with Natural Language
Processing (NLP) algorithms. In an impressive way, it engages users and interacts with them, answering their questions.
Conversation facilitators are mostly used by companies, government departments, and non-profit organisations. Money-related
industries such as banks, credit card companies, financial institutions, e-commerce stores, and startups are typical places where we
find these chatbots implemented. This research paper depicts the implementation and assessment of a Banking Conversational
Chatbot powered by Deep Learning (DL) techniques. The bank chatbot dataset, consisting of real user communication, was
preprocessed by cleaning, tokenisation, normalisation, and data balancing using SMOTE to ensure the training data was of the
highest quality. The authors proposed a Gated Recurrent Unit (GRU) network to capture the sequential dependencies and contextual
patterns of the user query, providing a more efficient and compact solution than the traditional LSTM model. In the conducted
comparative experiments with different models, namely SVM, XGBoost, and Naive Bayes, the accuracy recorded was 68%, 79%,
and 91%, respectively, while the argued GRU model results showed superiority over the other models with its accuracy of 97%,
precision of 97.9%, recall of 96%, and an F1-score of 97%. These figures demonstrate the GRU model's strength and effectiveness
in identifying user intent; thus, it can be a significant boost to the performance and reliability of conversational banking applications.