Büyük Dil Modeli Tabanli Veri Artirimi ile Türkçe Varlık İsmi Çıkarımı

Abstract

Named Entity Recognition (NER) plays a fundamental role in identifying and classifying named entities within texts. However, in resource-scarce languages and applications - particularly in Turkish - the lack of annotated data leads to a decline in model performance. In this study, synthetic examples were generated using Large Language Models (LLMs) to augment the existing primary dataset, with the aim of enhancing the k-shot learning performance of NER models. Experimental results demonstrate that models trained on the augmented dataset achieve performance improvements by a factor of 40 to 60 compared to those trained on the original dataset, indicating that the proposed method offers a cost-effective and viable alternative for resource-scarce applications. © 2025 Elsevier B.V., All rights reserved.