Implementation of the IndoBERT Model on Named Entity Recognition for Entity Identification in the Folklore of the Tale of Wayang Arjuna & Purusara

Abstract

The development of Natural Language Processing (NLP) has encouraged the use of Named Entity Recognition (NER) technology to extract important information from text data automatically. This research aims to apply the IndoBERT model to recognize named entities in classic Indonesian literary texts, namely Hikayat Wayang Arjuna & Purusara. The research focused on the identification of three categories of entities, namely Person (PER), Location (LOC), and Organization (ORG). The dataset is compiled through an annotation process using the BIO (Beginning, Inside, Outside) scheme, then processed through the preprocessing stage, tokenization using the IndoBERT Tokenizer, and fine-tuning the IndoBERT model. Evaluation was conducted using precision, recall, and F1-score metrics. The results showed that the IndoBERT model obtained a precision score of 56.30%, recall of 60.55%, and an F1-score of 58.35%. Based on the evaluation of each entity category, the Location (LOC) category obtained the best performance with an F1-score of 76.07%, followed by Person (PER) of 64.67%, while Organization (ORG) obtained a score of 44.96%. These results show that IndoBERT is quite effective in recognizing entities in classic literary texts in Indonesian and has the potential to support the process of information extraction, digitization, and preservation of folklore based on NLP technology.