gourishankar63/article-ner-v1

Article NER v1

Article NER v1 is a domain-specialized Named Entity Recognition (NER) model designed to identify and extract Person (PER) and Organization (ORG) entities from news articles.

Built on the DeBERTa-v3-small architecture, the model has been fine-tuned on a custom-curated newsroom dataset to recognize named entities in journalistic content.

The model follows the BIO (Beginning, Inside, Outside) tagging scheme to identify entity boundaries and classify entity mentions within article text.

Model Details

Attribute	Description
Model Name	Article NER v1
Model ID	gourishankar63/article-ner-v1
Base Architecture	DeBERTa-v3-small
Model Type	Token Classification
NLP Task	Named Entity Recognition
Training Dataset	Custom newsroom dataset
Annotation Scheme	BIO
Supported Entity Types	PER, ORG
Primary Domain	News Articles
License	Apache 2.0

Supported Entity Categories

The model currently recognizes two named entity categories:

Entity	Label	Description
Person	PER	Names of individuals and public figures
Organization	ORG	Companies, institutions, government bodies and other organizations

BIO Tagging Scheme

The model follows the standard BIO annotation convention:

• B-PER: Beginning of a person entity.
• I-PER: Inside a person entity.
• B-ORG: Beginning of an organization entity.
• I-ORG: Inside an organization entity.
• O: Token outside any recognized entity.

Training Dataset

The model was fine-tuned using a custom-built newsroom dataset consisting of article-based textual data.

The dataset was annotated using the BIO tagging scheme, focusing on identifying person and organization entities in journalistic contexts.

This domain-focused training helps the model recognize named entities within the linguistic and contextual patterns commonly found in news articles.

Evaluation Results

The model was evaluated on a separate held-out evaluation dataset.

The following entity-level F1 scores were obtained:

Entity Category	F1 Score
Person (PER)	97%
Organization (ORG)	91%

Performance Summary

• Person Recognition: 97% F1 score.
• Organization Recognition: 91% F1 score.

These results demonstrate strong entity recognition performance on the evaluated newsroom dataset.

Intended Use Cases

Article NER v1 is suitable for:

• Person and organization extraction from news articles.
• Automated newsroom content analysis.
• Article metadata enrichment.
• Entity-based article indexing and retrieval.
• News intelligence and information extraction.
• Knowledge graph construction.
• Entity-driven search and content discovery.

Usage

The model can be used with the Hugging Face Transformers library.

Installation

pip install transformers torch

Inference

from transformers import pipeline

ner_pipeline = pipeline(
    "token-classification",
    model="gourishankar63/article-ner-v1",
    aggregation_strategy="simple"
)

text = """
Microsoft CEO Satya Nadella announced a new
initiative in collaboration with OpenAI.
"""

entities = ner_pipeline(text)

for entity in entities:
    print({
        "entity": entity["entity_group"],
        "text": entity["word"],
        "confidence": round(entity["score"], 4)
    })

Example Output

[
    {
        "entity": "ORG",
        "text": "Microsoft",
        "confidence": 0.99
    },
    {
        "entity": "PER",
        "text": "Satya Nadella",
        "confidence": 0.98
    },
    {
        "entity": "ORG",
        "text": "OpenAI",
        "confidence": 0.99
    }
]

Note: Example predictions are illustrative and do not represent guaranteed inference scores.

Limitations

• The model currently focuses on Person (PER) and Organization (ORG) entities.
• It does not currently identify other entity types such as locations, dates, products, or events.
• Performance may vary across different writing styles, domains, and article formats.
• Ambiguous entity mentions and unfamiliar names may present challenges.
• Generalization to non-newsroom content has not been independently established.

License

This model is released under the Apache License 2.0, permitting use, modification, and distribution subject to the terms and conditions of the license.

Acknowledgements

This model builds upon the DeBERTa-v3-small pretrained architecture and has been fine-tuned using a custom newsroom dataset for domain-specific Named Entity Recognition.

Developer: gourishankar63

Model Repository: gourishankar63/article-ner-v1