Improving Sentence Similarity Estimation for Unsupervised Extractive Summarization

This repo contains the code, data and trained models for our ICASSP 2023 paper ''Improving Sentence Similarity Estimation for Unsupervised Extractive Summarization''

Requirements

• Python 3.7
• PyTorch
• Transformers

Install dependencies via:

conda create -n ss4sum python=3.8
conda activate ss4sum
pip install torch==1.9.0+cu111 torchvision==0.10.0+cu111 torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html
pip install -r requirements.txt

# For rouge-1.5.5.pl
sudo apt-get update
sudo apt-get install expat
sudo apt-get install libexpat-dev -y

sudo cpan install XML::Parser
sudo cpan install XML::Parser::PerlSAX
sudo cpan install XML::DOM

git clone https://github.com/summanlp/evaluation
pyrouge_set_rouge_path yourPath/evaluation/ROUGE-RELEASE-1.5.5

Download the Datasets and Best Checkpoint

Download processed datasets and checkpoint from https://drive.google.com/drive/folders/17wxORu-xmLPzGKVzecpikXaZeRkrjjmo?usp=sharing

The original datasets can be found at https://github.com/mswellhao/PacSum.

Train and Test

You may specify the hyper-parameters in exp/****.sh. We also provide the specific settings (train on CNNDM and NYT; test on CNNDM).

• Train:

bash exp/train.sh

• Test on CNNDM:

bash exp/test.sh