🎇AdaKGC

👋 News!

• Code for the paper Schema-adaptable Knowledge Graph Construction.

• Congratulations! Our work has been accepted by the EMNLP2023 Findings conference.

🎉 Quick Links

• Requirements
• Datasets of Extraction Tasks
• How to run
  • Named Entity Recognition Task
  • Relation Extraction Task
  • Event Extraction Task
• Inference
• Acknowledgment

🎈 Requirements

To run the codes, you need to install the requirements:

conda create -n adakgc python=3.8
pip install torch==1.8.0+cu111 -f https://download.pytorch.org/whl/torch_stable.html
pip install -r requirements.txt

🎏 Datasets of Extraction Tasks

Details of dataset construction see Data Construction.

You can find the dataset as following Google Drive links.

Dataset ACE05、Few-NERD、NYT

⚾ How to run

mkdir hf_models
cd hf_models
git lfs install
git clone https://huggingface.co/google/t5-v1_1-base
cd ..

mkdir output           # */AdaKGC/output

• Named Entity Recognition Task

# Current path:  */AdaKGC
. config/prompt_conf/Few-NERD_H.ini    # Load predefined parameters
bash scripts/run_finetune.bash --model=hf_models/t5-v1_1-base --data=data/Few-NERD_H/iter_1 --output=output/Few-NERD --mode=H --device=0 --batch=16

model: The name or path of a pre-trained model/checkpoint.

data: Path to the dataset.

output: The path of the fine-tuned checkpoint saved, the final automatically generated output path AdaKGC/output/ace05_event_H_e30_lr1e-4_b14_n0.8_prompt_80_800.

mode: segmentation mode (H、V、M or R).

device: CUDA_VISIBLE_DEVICES.

batch: batch size.

(see bash scripts and Python files for the detailed command-line arguments)

• Relation Extraction Task

. config/prompt_conf/NYT_H.ini  
bash scripts/run_finetune.bash --model=hf_models/t5-v1_1-base --data=data/NYT_H/iter_1 --output=output/NYT --mode=H --device=0 --batch=16

• Event Extraction Task

. config/prompt_conf/ace05_event_H.ini  
bash scripts/run_finetune.bash --model=hf_models/t5-v1_1-base --data=data/ace05_event_H/iter_1 --output=output/ace05_event --mode=H --device=0 --batch=16

🎰 Inference

• Only inference on single dataset(data/ace05_event_H/iter_1)

. config/prompt_conf/ace05_event_H.ini
CUDA_VISIBLE_DEVICES=0 python3 eval/inference.py --dataname=data/ace05_event_H/iter_1 --model=output/ace05_event_H_e30_lr1e-4_b16_n0.8_prompt_80_512 --task=event --cuda=0 --mode=H --use_ssi=${use_ssi} --use_task=${use_task} --use_prompt=${use_prompt} --prompt_len=${prompt_len} --prompt_dim=${prompt_dim}

datasetname: Path to the datasets which you want to predict.

• Automatic inference on all iteration datasets(data/iter_1/ace05_event_H ~ data/iter_7/ace05_event_H)

. config/prompt_conf/ace05_event_H.ini
CUDA_VISIBLE_DEVICES=0 python3 eval/inference_mul.py --dataname=ace05_event --model=output/ace05_event_H_e30_lr1e-4_b16_n0.8_prompt_80_512 --task=event --cuda=0 --mode=H --use_ssi=${use_ssi} --use_task=${use_task} --use_prompt=${use_prompt} --prompt_len=${prompt_len} --prompt_dim=${prompt_dim}

datasetname: datasets name(ace05_event、NYT or Few-NERD).

Complete process including fine-tune and inference (in scripts/run.bash):

mode=H
dataset_name=ace05_event
task=event
device=0
output_name=ace05_event_H_e30_lr1e-4_b16_n0.8_prompt_80_512
. config/prompt_conf/${dataset_name}_${mode}.ini 
bash scripts/run_finetune.bash --model=hf_models/mix --data=data/${dataset_name}_${mode}/iter_1 --output=output/${dataset_name} --mode=${mode} --device=${device} 
CUDA_VISIBLE_DEVICES=${device} python3 eval/inference_mul.py --dataname=${dataset_name} --model=${output_name} --task=${task} --mode=${mode} --use_ssi=${use_ssi} --use_task=${use_task} --use_prompt=${use_prompt} --prompt_len=${prompt_len} --prompt_dim=${prompt_dim}
CUDA_VISIBLE_DEVICES=${device} python3 eval/inference_mul.py --dataname=${dataset_name} --model=${output_name} --task=${task} --mode=${mode} --CD --use_ssi=${use_ssi} --use_task=${use_task} --use_prompt=${use_prompt} --prompt_len=${prompt_len} --prompt_dim=${prompt_dim}

Metric	Definition
ent-(P/R/F1)	Micro-F1 of Entity (Entity Type, Entity Span)
rel-strict-(P/R/F1)	Micro-F1 of Relation Strict (Relation Type, Arg1 Span, Arg1 Type, Arg2 Span, Arg2 Type)
evt-trigger-(P/R/F1)	Micro-F1 of Event Trigger (Event Type, Trigger Span)
evt-role-(P/R/F1)	Micro-F1 of Event Argument (Event Type, Arg Role, Arg Span)

🏳‍🌈 Acknowledgment

Part of our code is borrowed from UIE and UnifiedSKG, many thanks.

🚩 Papers for the Project & How to Cite

If you use or extend our work, please cite the paper as follows:

@article{DBLP:journals/corr/abs-2305-08703,
  author       = {Hongbin Ye and
                  Honghao Gui and
                  Xin Xu and
                  Huajun Chen and
                  Ningyu Zhang},
  title        = {Schema-adaptable Knowledge Graph Construction},
  journal      = {CoRR},
  volume       = {abs/2305.08703},
  year         = {2023},
  url          = {https://doi.org/10.48550/arXiv.2305.08703},
  doi          = {10.48550/arXiv.2305.08703},
  eprinttype    = {arXiv},
  eprint       = {2305.08703},
  timestamp    = {Wed, 17 May 2023 15:47:36 +0200},
  biburl       = {https://dblp.org/rec/journals/corr/abs-2305-08703.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}