Klasifikasi Named Entity Recognition (NER) Pada Cerita Pendek Bahasa Indonesia Menggunakan Model Indobert

Nazifa Samsurizal (Universitas Islam Negeri Ar-Raniry)
Hendri Ahmadian (UIN Ar-Raniry Banda Aceh)
Nurrizqa Nurrizqa (UIN Ar-Raniry Banda Aceh)

Progresif: Jurnal Ilmiah Komputer
Vol. 22 No. 3 (2026): Juli

Entity recognition in Indonesian short stories requires a model capable of understanding narrative language context effectively. This study aimed to implement the IndoBERT model for the Named Entity Recognition (NER) task to identify Person, Location, and Organization entities. The dataset was obtained from the Majalah Bobo e-book and organized using the BIO (Begin, Inside, Outside) format. The research included pre-processing, tokenization, label encoding, token alignment, and fine-tuning using the AutoModelForTokenClassification model. Evaluation was conducted using precision, recall, and F1-score metrics. The results showed that IndoBERT achieved F1-scores of 0.95 for Person, 0.81 for Organization, and 0.75 for Location, with a weighted average F1-score of 0.90. These results indicated that IndoBERT performed entity recognition effectively on Indonesian short stories. Keywords: Named Entity Recognition; IndoBERT; Indonesian short stories; Natural Language Processing; Transformer.    Abstrak Pengenalan entitas pada cerita pendek bahasa Indonesia memerlukan model yang mampu memahami konteks bahasa naratif secara efektif. Penelitian ini bertujuan menerapkan model IndoBERT pada tugas Named Entity Recognition (NER) untuk mengenali entitas Person, Location, dan Organization. Dataset diperoleh dari e-book Majalah Bobo dan disusun menggunakan format BIO (Begin, Inside, Outside). Penelitian dilakukan melalui tahapan pre-processing, tokenisasi, label encoding, token alignment, dan fine-tuning menggunakan model AutoModelForTokenClassification. Evaluasi dilakukan menggunakan metrik precision, recall, dan F1-score. Hasil pengujian menunjukkan bahwa model IndoBERT memperoleh nilai F1-score sebesar 0.95 pada label Person, 0.81 pada Organization, dan 0.75 pada Location, dengan weighted average F1-score sebesar 0.90. Hasil tersebut menunjukkan bahwa IndoBERT mampu melakukan pengenalan entitas dengan baik pada cerita pendek bahasa Indonesia. Kata kunci: Named Entity Recognition; IndoBERT; cerita pendek bahasa Indonesia; Natural Language Processing; Transformer.