Name Entity Recognition (NER) Dataset

The Name Entity Recognition (NER) Dataset is an annotated text dataset designed for natural language processing and named entity recognition tasks. It is derived from the Groningen Meaning Bank (GMB) corpus and contains 47,959 sentences with approximately 1.35 million words.

Each word receives a corresponding entity tag, allowing machine learning models to learn how to identify important entities within text. In addition, the dataset includes part-of-speech information, which can provide an additional linguistic feature during model development.

The dataset follows a token-level annotation structure. Therefore, developers can group individual words into sentences and train sequence-labeling models for NER applications.

Key Features

• 47,959 annotated sentences

• Approximately 1.35 million words

• Word-level named entity annotations

• Sentence identifiers for grouping tokens into sentences

• Part-of-speech (POS) tags

• Named entity recognition (NER) tags

• Suitable for sequence-labeling tasks

• Useful for NLP and information extraction projects

• Supports machine learning and deep learning experiments

Data Included in the NER Dataset

The dataset contains four primary columns:

The Sentence # column allows users to reconstruct complete sentences from individual token rows. Meanwhile, Word contains the actual text, POS provides grammatical information, and Tag identifies the named entity category.

Named Entity Categories

The dataset contains multiple entity categories that represent different types of information in text. These include:

• Geographical entities (geo)

• Organizations (org)

• Persons (per)

• Geopolitical entities (gpe)

• Time indicators (tim)

• Artifacts (art)

• Events (eve)

• Natural phenomena (nat)

The annotation scheme also uses beginning and inside tags to represent multi-word entities. For example, B- identifies the beginning of an entity while I- identifies a subsequent token within the same entity. The O tag represents tokens that do not belong to a named entity.

Applications of the Name Entity Recognition Dataset

Named Entity Recognition

Train NLP models to identify people, organizations, locations, dates, and other entities within unstructured text.

Information Extraction

Extract structured information from documents, articles, reports, and other text sources.

Natural Language Processing

Use the annotated text to develop and evaluate NLP pipelines for entity identification and text understanding.

Sequence Labeling

The token-level structure makes the dataset suitable for sequence-labeling models such as CRF, BiLSTM, and transformer-based architectures.

Text Classification

Combine entity information with other linguistic features to improve text classification and document analysis workflows.

Machine Learning and Deep Learning

The dataset can support experiments with traditional machine learning algorithms as well as deep learning models for NLP. Researchers have used the dataset for models involving BiLSTM, CRF, and other sequence-processing approaches.

Why Use the NER Dataset?

The dataset provides a large collection of annotated text for practical NER experiments. Moreover, its word-level labels allow models to learn how entity information appears within sentences rather than relying only on document-level classifications.

The inclusion of POS tags provides additional linguistic information. As a result, researchers can experiment with different feature combinations when building NER systems.

Furthermore, the dataset covers several entity types, including people, organizations, geographical locations, time expressions, events, and artifacts. Therefore, it can support a range of information extraction and NLP applications.

Who Can Use This Dataset?

The Name Entity Recognition Dataset can be useful for:

• NLP researchers

• Machine learning developers

• Data scientists

• AI researchers

• NLP students

• Deep learning practitioners

• Information extraction developers

• Natural language processing teams

• Researchers building sequence-labeling models

Conclusion

The Name Entity Recognition (NER) Dataset provides a substantial collection of annotated text for developing and evaluating NLP models. With approximately 47,959 sentences, around 1.35 million words, POS information, and entity-level tags, it offers a useful resource for named entity recognition and information extraction.

Researchers and developers can use the dataset to experiment with sequence-labeling models, NLP pipelines, and machine learning approaches for identifying entities in unstructured text.

ThisDataset is available on Kaggle.