AgaCKNER: First Kurdish Sorani Named Entity Recognition Dataset

AgaCKNER is the first publicly accessible Named Entity Recognition (NER) dataset in the Kurdish Sorani language, developed to advance research in low-resource language processing. Derived from the Rudaw Media Network, AgaCKNER encompasses a broad array of topics across five distinct domains: Kurdistan news, Middle East news, world news, economic news, and sports news that are meticulously curated from over 160 articles. The dataset includes 2,534 sentences and 64,563 tokens, pre-processed and formatted in CoNLL for NER tasks. Entities are labelled in BIO format under five categories: PERSON, LOCATION, ORGANIZATION, DATE, and Miscellaneous. AgaCKNER is an essential resource for Kurdish Sorani natural language processing, greatly enhancing research in low-resource languages. Its structure makes it easily adaptable for generating training, validation, and test splits.

AgaCKNER是首个可公开获取的库尔德索拉尼语（Kurdish Sorani）命名实体识别（Named Entity Recognition, NER）数据集，旨在推动低资源语言处理领域的研究。该数据集源自鲁多媒体网络（Rudaw Media Network），从160余篇文章中精心筛选出覆盖五大领域的广泛主题，具体包括库尔德斯坦新闻、中东新闻、世界新闻、经济新闻与体育新闻。数据集共包含2534个句子与64563个Token，已完成预处理并采用CoNLL格式进行格式化，以适配命名实体识别任务。实体采用BIO标注格式，分为五大类别：人物（PERSON）、地点（LOCATION）、组织机构（ORGANIZATION）、日期（DATE）与杂项（Miscellaneous）。AgaCKNER是库尔德索拉尼语自然语言处理的核心资源，极大推动了低资源语言领域的研究工作。其结构具备良好的灵活性，可便捷地划分为训练集、验证集与测试集。