Penerapan Named Entity Recognition (NER) Untuk Ekstraksi Otomatis Entitas Pada Teks Berita Online Pemerintah Daerah

Abstrak

Local government news is an important source of public information, but it is generally presented in unstructured text, making it difficult to search and analyze. This study aims to apply Named Entity Recognition (NER) to extract key entities and develop a web-based system that integrates news scraping, text preprocessing, entity extraction, and structured result presentation. The method utilizes a multilingual pretrained spaCy model and an IndoBERT pretrained model to recognize entities such as PER (person), ORG (organization), LOC (location), and DATE in news from Detik.com. The system was developed using the Flask framework and evaluated by comparing extraction results with ground truth data using precision, recall, and F1-score metrics. The evaluation results show a precision of 67%, recall of 76.2%, and an F1-score of 71.2%. The best performance was achieved for PER and DATE entities, while LOC and ORG entities showed less optimal results. Overall, this system can be used as an initial tool to support the analysis of local government news more efficiently and systematically.