Graph Neural Networks with Generated Parameters for Relation Extraction

Code and dataset of ACL2019 Paper Graph Neural Networks with Generated Parameters for Relation Extraction. paper

Citation

Please cite our paper if you find it helpful.

@inproceedings{zhu2019graph,
  title={Graph Neural Networks with Generated Parameters for Relation Extraction},
  author={Zhu, Hao and Lin, Yankai and Liu, Zhiyuan and Fu, Jie and Chua, Tat-Seng and Sun, Maosong},
  booktitle={Proceedings of ACL},
  year={2019}
}

This repository is mainly contributed by Hao Zhu, Chaojun Xiao. And this repository is developed based on ContextAware.

Dataset

Please download gpgnn_data.zip from Aliyun and unzip it as the data folder:

unzip gpgnn_data.zip

Requirements and Installation

This repository has been tested with Python 3.6,torch==1.4.0,sacred==0.8.1

pip3 install -r requirements.txt

Get Started

Directory structure

models/: Baseline models (LSTM, CNN, PCNN, ContextAware) in baselines.py and GPGNN model in our_models.py.

parsing/: APIs to convert graphs into tensors which can be fed into our models.

semanticgraph/: APIs to construct relation graphs from sentences.

utils/: APIs to load word embeddings, evaluate, and operate the graphs.

result/: Storage area for models and output results on the test set.

model_param.json: Hyper parameters for the GPGNN model.

Running

train:

The parameters of the function main are set in function main_config. You can modify the parameters in main_config to train different models. And the hyper parameters for the models are stored in model_params.json.

python3 train.py

test:

Before testing, you should choose proper version of the model by modifing the load_model in main_config.

python3 test.py

evaluate:

cd result
python3 test_accuracy.py --input (result_file_path)

thunlp/GP-GNN