KeyEE: Enhancing Low-resource Generative Event Extraction with Auxiliary Keyword Sub-Prompt

Code repository for paper "KeyEE: Enhancing Low-resource Generative Event Extraction with Auxiliary Keyword Sub-Prompt".

Our code is mainly based on DEGREE. We deeply thank the contribution from the authors of the paper.

Environment

• torch==1.8.0
• transformers==4.25.1
• protobuf==3.20.3
• tensorboardx==2.6
• lxml==4.9.2
• beautifulsoup4==4.11.2
• bs4==0.0.1
• stanza==1.4.2
• ipdb==0.13.11

conda create -n keyee python=3.8
conda activate keyee
python -m pip install -r requirements.txt

Datasets

We support ace05e, ace05ep, and ere.

Preprocessing

Our preprocessing mainly adapts DEGREE's and OneIE's released scripts with minor modifications. We deeply thank the contribution from the authors of the paper. We propose to create a new virtual environment to complete the data preprocessing.

ace05e

1. Prepare data processed from DyGIE++
2. Put the processed data into the folder processed_data/ace05e_dygieppformat
3. Run ./scripts/process_ace05e.sh

ace05ep

1. Download ACE data from LDC
2. Run ./scripts/process_ace05ep.sh

ere

1. Download ERE English data from LDC, specifically, "LDC2015E29_DEFT_Rich_ERE_English_Training_Annotation_V2", "LDC2015E68_DEFT_Rich_ERE_English_Training_Annotation_R2_V2", "LDC2015E78_DEFT_Rich_ERE_Chinese_and_English_Parallel_Annotation_V2"
2. Collect all these data under a directory with such setup:

ERE
├── LDC2015E29_DEFT_Rich_ERE_English_Training_Annotation_V2
│     ├── data
│     ├── docs
│     └── ...
├── LDC2015E68_DEFT_Rich_ERE_English_Training_Annotation_R2_V2
│     ├── data
│     ├── docs
│     └── ...
└── LDC2015E78_DEFT_Rich_ERE_Chinese_and_English_Parallel_Annotation_V2
      ├── data
      ├── docs
      └── ...

3. Run ./scripts/process_ere.sh

The above scripts will generate processed data (including the full training set and the low-resourece sets) in ./process_data.

Training

All training configurations are listed in config directory, you should check your configurations before experiments.

Run ./scripts/train.sh or use the following commands:

Generate data

python keyee/generate_data.py -c config/config_keyee_ace05e.json

Train

python keyee/train.py -c config/config_keyee_ace05e.json

Evaluation

We negatively sampled those sentences that were missing a certain event type during the training phase to reduce training time, which means we did not retrain full dev and test dataset in training stage. So it is important to do extra evaluation on the whole test datset.

To do this, you can run ./scripts/eval.sh or use the following commands:

python keyee/eval.py \
    -c config/config_keyee_ace05e.json \
    -e $OUTPUT_DIR/best_model.mdl \
    --eval_batch_size 16 \
    --write_file $OUTPUT_DIR/eval_result.json \
    --no_dev

Citation

If you find that the code is useful in your research, please consider citing our paper.

@ARTICLE{BDMA2024_KeyEE,
    author  = {Duan, Junwen and Liao, Xincheng and An, Ying and Wang, Jianxin},
    journal = {Big Data Mining and Analytics}, 
    title   = {KeyEE: Enhancing Low-Resource Generative Event Extraction with Auxiliary Keyword Sub-Prompt}, 
    year    = {2024},
    volume  = {7},
    number  = {2},
    pages   = {547-560},,
    doi     = {10.26599/BDMA.2023.9020036}
}

Contact

If you have any issue, please contact Xincheng Liao at (ostars@csu.edu.cn)