<![CDATA[Named Entity Recognition (NER) is a fundamental task in natural language processing that supports information extraction and knowledge organization. However, empirical studies examining the computational scalability of conventional NER models for the Indonesian language remain limited. This study investigates the scalability and computational performance of the BiLSTM-CRF model for Indonesian NER tasks. The objective is to evaluate how the model’s computational requirements and predictive performance change as the size of the training dataset increases. An experimental evaluation was conducted by training the BiLSTM-CRF model on three dataset scales derived from the WikiANN corpus (small, medium, and large) using a standard configuration with randomly initialized embeddings in a CPU-based environment. Model performance was assessed using the F1-score, while computational scalability was analyzed through measurements of training time, memory consumption, and inference speed. The results indicate a clear scalability pattern in which computational costs increase with dataset size, particularly in training time and memory usage. At the same time, predictive performance improves as more training data becomes available, with the F1-score increasing from 0.70 on the smallest dataset to 0.86 on the largest dataset. These findings provide empirical evidence on the scalability behavior of the BiLSTM-CRF model for Indonesian NER and offer practical insights for selecting model configurations under limited computational resources. ]]>
Copyrights © 2026
