RNN Based Customer Service Chatbot for Information Support at Gleamore

Abstract

In the rapidly evolving digital era, PT.Gleamore Digital Solution faces challenges in managing a high volume of customer inquiries, often leading to response delays that affect service quality and customer satisfaction. This study aims to design and develop a customer service chatbot using the Recurrent Neural Network (RNN) method to support automated and efficient information services. The research involves several key stages, including data preprocessing (tokenization, stemming, case folding, and bag of words), model training, and evaluation. The dataset, obtained from company administrators, consists of frequently asked questions categorized into multiple service-related intents. Experimental results show that the RNN model outperformed the combined RNN-LSTM method, achieving a validation accuracy of 93.55% with a final loss of 0.2908 and faster training time. Model evaluation using a confusion matrix achieved an accuracy rate of 98.70%, indicating high reliability in intent recognition and response generation. Black Box Testing also confirmed that the chatbot consistently provided valid and relevant responses to various user queries. The results imply that the RNN-based chatbot effectively improves the efficiency of customer service operations, reduces administrative workload, and enhances user satisfaction by enabling accurate and real-time responses through automated digital interaction.