luke_el

A library for luke base bi-encoding entity linking models

Usage

Instllation

git clone [email protected]:YuSawan/luke-el.git
cd luke_el
python -m venv .venv
source .venv/bin/activate
pip install .

Dataset preparation

Dataset

{
  "id": "doc-001",
  "examples": [
    {
      "paragraph-id": "doc-001-P1",
      "text": "She graduated from NAIST.",
      "entities": [
        {
          "start": 19,
          "end": 24,
          "label": ["000011"]
        }
      ],
    }
  ]
}

Dictionary

{
  "id": "000011",
  "name": "NAIST",
  "description": "NAIST is located in Ikoma."
}

Finetuning

1st stage

python luke_el/cli/run.py \
    --do_train \
    --config_file configs/config.yaml \
    --output_dir ./initial_output/

2nd stage

python luke_el/cli/run.py \
    --do_train \
    --config_file configs/config.yaml \
    --negative dense \
    --prev_path ./initial_output/ \
    --output_dir ./second_output/

Evaluation/Prediction

python luke_el/cli/run.py \
    --do_eval \
    --do_predict \
    --config_file configs/config.yaml \
    --prev_path PATH_TO_YOUR_MODEL \
    --output_dir PATH_TO_YOUR_MODEL