Analisis Perbandingan Klasifikasi Intent Chatbot Menggunakan Deep Learning BERT, RoBERTa, dan IndoBERT

A chatbot is a software application to designed handle user inputs and generate appropriate replies based on those inputs, which are then communicated back to the user. In able to provide accurate responses, the chatbot must be able to understand the intent of the user accurately. An issue in the development of chatbots is how to accurate classify user intent. Incorrectly understanding user intent can result in irrelevant responses. In order to have a conversation with the user, the intent of the user needs to be classified correctly. This paper compares three state-of-the-art transformer-based models BERT (Bidirectional Encoder Representations from Transformers), RoBERTa (Robustly Optimized BERT Pretraining Approach), and IndoBERT (Indonesia Bidirectional Encoder Representations from Transformer) for the task of intent classification in chatbot systems. Various performance metrics, including accuracy, F1-score, precision, and recall, were analyzed to determine which model performs more effectively in the same parameter conditions. Performance metrics like accuracy and F1-score were compared to assess model BERT, RoBERTa and IndoBERT performs better in a University Chatbot Dataset in Indonesian language. The BERT model achieved an accuracy of 0.89, RoBERTa model achieved 0.84 and IndoBERT model achieved an accuracy of 0.94. The better performance of IndoBERT compared to BERT and RoBERTa is caused by more language-specific training, more relevant pretraining, and more effective adaptation to Indonesian context and structure.