Skip to content

Latest commit

 

History

History
126 lines (93 loc) · 3.23 KB

README.md

File metadata and controls

126 lines (93 loc) · 3.23 KB
Raw

Model Description

This model is a fine-tuned version of the tohoku-nlp/bert-base-japanese-v3, specifically optimized for Named Entity Recognition (NER) tasks. It is fine-tuned using a Japanese named entity extraction dataset derived from Wikipedia, which was developed and made publicly available by Stockmark Inc. (NER Wikipedia Dataset).

Intended Use

This model is intended for use in tasks that require the identification and categorization of named entities within Japanese text. It is suitable for various applications in natural language processing where understanding the specific names of people, organizations, locations, etc., is crucial.

How to Use

Download Model Weights/Files

Get the model weights from huggingface and save it to ner_model directory. In future I will make this process more straightforward

cd ner_model
wget https://huggingface.co/knosing/japanese_ner_model/resolve/main/model.safetensors
wget https://huggingface.co/knosing/japanese_ner_model/resolve/main/training_args.bin
wget https://huggingface.co/knosing/japanese_ner_model/resolve/main/config.json

Local Development

Conda

Create a conda virtual environment and activate it:

conda create --name ner_fa python=3.10
conda activate ner_fa
pip install -r requirements.txt --no-cache-dir
export PYTHONPATH="$PWD"

Run Streamlit App (if you have model weights, else train model first)

streamlit run app.py

Training

If you wish to train the NER model

python3 app/services/model.py

Docker

Train the NER model and save it to ner_model folder. It should contain the following files:

  • config.json
  • model.safetensors
  • training_args.bin

Create image (will take about 5min)

docker build . -t ner

Run

docker run -p 8080:8080 ner

Result

The model has been evaluated on various entity types to assess its precision, recall, F1 score, and overall accuracy. Below is the detailed performance breakdown by entity type:

Overall Metrics

  • Overall Precision: 0.8379
  • Overall Recall: 0.8477
  • Overall F1 Score: 0.8428
  • Overall Accuracy: 0.9684

Performance by Entity Type

  • Other Organization Names (の他の組織名):

    • Precision: 0.71875
    • Recall: 0.69
    • F1 Score: 0.7041
    • Sample Count: 100

  • Event Names (ベント名):

    • Precision: 0.85
    • Recall: 0.8586
    • F1 Score: 0.8543
    • Sample Count: 99

  • Personal Names (人名):

    • Precision: 0.8171
    • Recall: 0.8664
    • F1 Score: 0.8410
    • Sample Count: 232

  • Generic Names (名):

    • Precision: 0.8986
    • Recall: 0.9376
    • F1 Score: 0.9177
    • Sample Count: 529

  • Product Names (品名):

    • Precision: 0.6522
    • Recall: 0.5906
    • F1 Score: 0.6198
    • Sample Count: 127

  • Government Organization Names (治的組織名):

    • Precision: 0.9160
    • Recall: 0.8276
    • F1 Score: 0.8696
    • Sample Count: 145

  • Facility Names (設名):

    • Precision: 0.7905
    • Recall: 0.8357
    • F1 Score: 0.8125
    • Sample Count: 140

Notes

Have any issues? Please feel free to contact me.