<![CDATA[Fragmented school communication can delay access to official information and increase repetitive administrative inquiries. This study developed and evaluated an intelligent school-information website integrated with an Indonesian-language chatbot at MA Islamiyah Bindu. The web application was implemented using PHP, Bootstrap 5, and MySQL, while the chatbot employed the indobenchmark/indobert-base-p1 model for multi-class intent classification. A balanced dataset of 600 school-related utterances covering ten intents was prepared and divided into 500 training and 100 independent test samples. IndoBERT was fine-tuned for 40 epochs using a maximum sequence length of 64 and a batch size of 8. Training loss decreased from 2.4970 to 0.0027, while training accuracy reached 100%. On the held-out test set, the model achieved 94.0% accuracy, 94.3% macro-precision, 94.0% macro-recall, and 94.0% macro-F1. Black-box testing confirmed that all principal website modules, chatbot interactions, controlled response retrieval, and fallback handling operated correctly. The integrated architecture provided accurate, responsive, and centrally managed access to school profiles, schedules, facilities, admission requirements, and contact information.]]>
Copyrights © 2026
