kaiserahmed/BanglaTag-NER-Dataset

BanglaTag NER数据集是一个孟加拉语（Bangla）命名实体识别数据集，包含22,144个句子，使用BIO标记方案标注了13种实体类型。数据集格式为每行一个JSON对象，包含tokens和ner_tags字段。实体类型包括人名（PER）、地点（LOC）、组织（ORG）、政治实体（POL）、日期（DATE）、时间（TIME）、事件（EVENT）、犯罪相关实体（CRIME）、头衔（TITLE）、数字（NUM）、符号（SYMBOL）、选区（CONSTITUENCY）和机构（INST）。数据集统计信息显示总句子数为22,144，其中训练集占85%（约18,822句），验证集和测试集各占7.5%（约1,661句）。数据集还提供了标签列表和使用示例，包括通过HuggingFace的load_dataset函数加载或手动加载JSONL文件的方法。此外，README还提到了一个基于该数据集微调的NER模型，并提供了引用信息。

The BanglaTag NER Dataset is a Bengali (Bangla) Named Entity Recognition dataset containing 22,144 sentences labeled using the BIO tagging scheme across 13 entity types. Each line in the dataset is a JSON object with tokens and ner_tags fields. The entity types include PER (person names), LOC (locations, cities, countries), ORG (organizations, companies), POL (political entities/parties), DATE (calendar dates), TIME (times of day), EVENT (named events), CRIME (crime-related entities), TITLE (titles, designations), NUM (numbers, quantities), SYMBOL (symbols, currencies), CONSTITUENCY (electoral constituencies), and INST (institutions). The dataset statistics show a total of 22,144 sentences, with 85% (~18,822) in the training set, and 7.5% each (~1,661) in the validation and test sets. The README also provides a label list, usage examples (including loading via HuggingFaces load_dataset function or manually loading JSONL files), and mentions a fine-tuned NER model based on this dataset. Citation information is also included.