Fine-grained Named Entity Recognition for Turkish

Abstract

Named Entity Recognition is the process of finding and labelling entity names in text. It is also important to use fine-grained entities in different application areas such as relation extraction, knowledge base building and question answering. However, studies in Turkish are insufficient in this context. In order to overcome this deficiency, we created large-scale dataset with 23 different fine-grained entities for Turkish. These fine-grained entities consist of subcategories of people, institutions and place entities. In the article, experiments were conducted with the transformer-based models in Turkish. While a F1 score of 93.34%, which is close to the state-of-the-art for Turkish, was obtained for general entity names, the performance for subcategories was 79.82%.