## KazNERD: Kazakh Named Entity Recognition Dataset

**Dataset Summary:**

The KazNERD dataset provides a valuable resource for Kazakh Named Entity Recognition (NER).  It comprises 112,702 sentences extracted from television news text, manually annotated by two native Kazakh speakers under expert supervision.  The dataset uses the IOB2 tagging scheme and features 136,333 annotations across 25 entity classes.  Accompanying annotation guidelines (in Kazakh) and code for training various NER models (CRF, BiLSTM-CNN-CRF, BERT, and XLM-RoBERTa) are also publicly available.

**Dataset Details:**

* **Source:** Television news text.
* **Size:** 112,702 sentences, 136,333 annotations.
* **Annotation scheme:** IOB2
* **Number of entity classes:** 25
* **License:** CC BY 4.0
* **Repository:** [https://github.com/IS2AI/KazNERD](https://github.com/IS2AI/KazNERD)


**Languages:**

* Kazakh


**Data Fields:**

The data is provided in the CoNLL 2002 format.  Specific field descriptions are available within the dataset files.


**Additional Information:**

The dataset includes annotation guidelines (in Kazakh) clarifying the annotation process.  The repository also contains code for training NER models using different architectures, allowing for easy replication of the reported results.