<![CDATA[The development of Artificial Intelligence (AI) technology in education provides convenience in the learning process, but also creates challenges related to understanding the ethics of AI use in academic environments. This study aims to evaluate the performance of several BERT model variants in performing intent recognition for an educational chatbot on the ethical use of AI, particularly in higher education environments. The models used include IndoBERT, Multilingual BERT, and DistilBERT. The dataset consists of 700 data points with 7 intent categories developed using a Retrieval-Augmented Generation (RAG) approach based on the 2024 guideline book on the use of Generative AI from the Ministry of Education, Culture, Research, and Technology. The models were evaluated using accuracy, precision, recall, and F1-score metrics, while also handling out-of-scope (OOS) questions by comparing confidence threshold and entropy-based detection methods. The results show that IndoBERT achieved the best performance, with accuracy, precision, recall, and F1-score values of 97 percent, outperforming Multilingual BERT and DistilBERT. In addition, the entropy-based detection method achieved an accuracy of 95 percent and performed better in detecting out-of-scope questions compared to the confidence threshold method. These findings indicate that IndoBERT is an effective model for intent recognition in an educational chatbot on the ethical use of AI in academic environments.]]>
Copyrights © 2026
