bert-ner-finetune

BERT NER Fine-tuning

Fine-tune BERT for Named Entity Recognition (NER) using HuggingFace and the CoNLL-2003 dataset.

---

Structure

• notebooks/ner_finetune.ipynb – Main notebook for training + evaluation
• notebooks/ner_visualize.ipynb – Visualize model predictions on custom text
• train.py – Script to fine-tune BERT
• infer.py – Inference script for user-defined text
• evaluate.py – Evaluation script with F1-score and classification report
• data/ – Holds datasets or preprocessed splits
• outputs/ – Stores saved models or predictions

---

Goals

• Load and preprocess CoNLL-2003 dataset
• Fine-tune BERT for token classification
• Evaluate with precision, recall, F1
• Visualize predictions in plain text and color-coded HTML

---

Models

• bert-base-cased (used in CoNLL-2003 benchmarks)

---

Setup

pip install -r requirements.txt

---

Dependencies

• transformers
• datasets
• seqeval
• torch
• pandas
• matplotlib
• scikit-learn