SAPGraph

This repository contains code and data for paper SAPGraph: Structure-aware Extractive Summarization for Scientific Papers with Heterogeneous Graph.

Environment

For pyTorch and DGL, I use

torch==1.9.0+cu111
dgl==1.1.2+cu111

Should install the right dgl package for your system and cuda, through https://www.dgl.ai/pages/start.html. Note that dgl-0.x.x is a bit differet with dgl-1.x.x.

Run

pip install -r requirements.txt

Data Prepare

Raw data is avaliable on Google Drive: https://drive.google.com/file/d/1_5gZkOFYeB4FKrt-60J7bxC09iULAniI/view?usp=sharing This includes the source document (parsed from paper pdf) and its summary as a pair.

After putting the data into dir data/cordSum, you need to run PrepareDataset.sh to prepare graph meta data for each sample, including node and edge construction. And the result will save into dir cache/cordSum.

bash PrepareDataset.sh cordSum cord/cordSum

Training

Some code are borrowed from HSG. Thanks for their work.

bash train_cord.sh

Please make sure the settings is what you want before running.

Testing

bash test_cord.sh