Development of an Intent-Classification Chatbot to Support Operational Services at Kadin Indonesia

Abstract

The digital transformation era demands business membership organizations such as the Indonesian Chamber of Commerce and Industry (Kadin) to provide responsive and scalable services. Operational inquiries related to the Certificate of Origin (COO), membership information (KTA), activity agendas, and administrative correspondence are still predominantly handled manually, resulting in service queues and limited operating hours. This study develops an intelligent text-based chatbot using Natural Language Processing (NLP) with an intent classification approach implemented through a Long Short-Term Memory (LSTM) model to automate initial responses to user queries. A labeled dataset consisting of more than 90 intents was constructed from Frequently Asked Questions (FAQ), Kadin service data, and data augmentation to increase text variation. The preprocessing pipeline includes normalization, tokenization, padding, and 300 dimensional FastText embeddings. The LSTM model, configured with 128 units, was trained using categorical cross-entropy with a label smoothing factor of 0.05, the Adam optimizer, a batch size of 20, and 80 epochs, and integrated into the backend for real-time inference. Evaluation on the test set achieved an accuracy of 92.08% and a Top-3 Accuracy of 96.23%. Visual analyses using the confusion matrix and accuracy–loss curves indicate strong generalization capability. These findings demonstrate that a properly configured LSTM model can effectively recognize service-related intents for Kadin.