---
language: en
tags:
  - token-classification
  - named-entity-recognition
  - ner
  - bert
  - conll2003
datasets:
  - conll2003
metrics:
  - seqeval
model-index:
  - name: named_entity-recognition
    results:
      - task:
          type: token-classification
        dataset:
          name: CoNLL-2003
          type: conll2003
        metrics:
          - type: f1
            value: 0.9116
          - type: precision
            value: 0.9041
          - type: recall
            value: 0.9192
---

# BERT Fine-Tuned for Named Entity Recognition (CoNLL-2003)

This model recognizes named entities in English text: **People**, **Organizations**,
**Locations**, and **Miscellaneous** entities.

## Model Details

- **Base model:** bert-base-cased
- **Dataset:** CoNLL-2003 (14,041 training sentences from Reuters news)
- **Task:** Named Entity Recognition (token classification)
- **Framework:** PyTorch + HuggingFace Transformers

## Entity Types

| Label | Meaning | Example |
|-------|---------|---------|
| PER | Person names | Barack Obama, Elon Musk |
| ORG | Organizations | Apple Inc., United Nations |
| LOC | Locations | New York, Mount Everest |
| MISC | Miscellaneous | English, FIFA World Cup |

## Performance (CoNLL-2003 Test Set)

| Metric | Score |
|--------|-------|
| F1 Score | 0.9116 |
| Precision | 0.9041 |
| Recall | 0.9192 |
| Accuracy | 0.9827 |

## How to Use

```python
from transformers import pipeline

# Load the model
ner = pipeline(
    "token-classification",
    model="samandar1105/named_entity-recognition",
    aggregation_strategy="simple"
)

# Run inference
result = ner("Elon Musk founded SpaceX in Hawthorne, California.")
print(result)
# [
#   {'entity_group': 'PER', 'word': 'Elon Musk', 'score': 0.998},
#   {'entity_group': 'ORG', 'word': 'SpaceX', 'score': 0.997},
#   {'entity_group': 'LOC', 'word': 'Hawthorne', 'score': 0.995},
#   {'entity_group': 'LOC', 'word': 'California', 'score': 0.994},
# ]
```

## Training Details

- Learning rate: 2e-5
- Epochs: 4
- Batch size: 16
- Max sequence length: 128
- Warmup ratio: 0.1
- Weight decay: 0.01
- Label alignment: First-subword strategy with -100 for continuation subwords
- Evaluation: seqeval (entity-level strict span matching)