BERT NER

Use google BERT to do CoNLL-2003 NER !

Requirements

• python3

• pip3 install -r requirements.txt

Run

python run_ner.py --data_dir=data/ --bert_model=bert-base-cased --task_name=ner --output_dir=out --max_seq_length=128 --do_train --num_train_epochs 5 --do_eval --warmup_proportion=0.4

Result

Validation Data

             precision    recall  f1-score   support

       MISC     0.9407    0.9304    0.9355       273
        LOC     0.9650    0.9881    0.9764       419
        PER     0.9844    0.9783    0.9813       322
        ORG     0.9794    0.9852    0.9822       337

avg / total     0.9683    0.9734    0.9708      1351

Test Data

             precision    recall  f1-score   support

        ORG     0.9152    0.9073    0.9113       464
        PER     0.9767    0.9692    0.9730       260
        LOC     0.9397    0.9263    0.9330       353
       MISC     0.8276    0.9014    0.8629       213

avg / total     0.9198    0.9240    0.9217      1290

Pretrained model download from here

Inference

from bert import Nermodel = Ner("out/")output = model.predict("Steve went to Paris")print(output)# {#     "Steve": {#         "tag": "B-PER",#         "confidence": 0.999879002571106#     },#     "went": {#         "tag": "O",#         "confidence": 0.9968552589416504#     },#     "to": {#         "tag": "O",#         "confidence": 0.9996656179428101#     },#     "Paris": {#         "tag": "B-LOC",#         "confidence": 0.999504804611206#     }# }

Tensorflow version

• https://github.com/kyzhouhzau/BERT-NER