Named Entity Recognition on Indonesian Online News Based on Bidirectional LSTM-CRF

The rapid growth of information in Indonesian online news has created an urgent need for effective methods to automatically identify and classify critical entities. Named Entity Recognition (NER) plays a vital role in natural language processing by extracting entities such as persons, organizations, locations, and time expressions. However, challenges such as the flexible linguistic structure of the Indonesian language, lack of high-quality annotated datasets, and the prevalence of informal language pose significant obstacles. This study proposes a Named Entity Recognition model based on a Bidirectional Long Short-Term Memory with a Conditional Random Field layer (Bi-LSTM-CRF) to enhance entity recognition accuracy in Indonesian news texts. The dataset consists of 47,955 annotated sentences, and the study compares the impact of two word embedding techniques: Continuous Bag of Words (CBOW) and Skip-gram. Experimental results demonstrate that the Skipgram + Bi-LSTM-CRF model outperforms others, achieving an F1-score of 0.8211. This model significantly improves over the baseline Bi-LSTM and previously studied Bi-LSTM-CNN models. The key contributions of this research include: (1) demonstrating the effectiveness of CRF in improving sequence label predictions in the Indonesian NER task, (2) evaluating the impact of different word embeddings on model performance, and (3) providing a robust framework for automatic information extraction in the Indonesian digital news domain. These findings have potential applications in automatic news summarization, intelligent search systems, and entity-based sentiment analysis for the Indonesian language.