This repository contains the source code of the paper "A Hierarchical Entity Graph Convolutional Network for Relation Extraction across Documents" published in RANLP 2021.
The dataset used for experiments in the paper can be downloaded from the following link:
https://drive.google.com/drive/folders/1nBima4wLXSRHtMfcMdcQgu3AFlMe9AWP?usp=sharing
- python3.6
- pytorch 1.7
- CUDA 8.0
python3.6 models.py source_dir embedding_file target_dir model_id train
embedding_file = path for glove.6B.300d.txt
python3.6 models.py source_dir embedding_file target_dir model_id test threshold
Above threshold value will be one of outputs of the training process
Use model_id as 1 for CNN, 2 for BiLSTM, 3 for BiLSTM_CNN, 4 for LinkPath, and 5 for our HEGCN model.