Lay Summarization with SummerNet on ELIFE AND PLOS Dataset

Quick Links

• Environment
• Data Preprocessing
• Train
• Inference

Environment

Create the environment with conda and pip.

conda env create -f environment.yml
conda activate season
pip install -r requirements.txt

Install nltk "punkt" package.

python -c "import nltk; nltk.download('punkt');"

We've tested this environment with python 3.8 and cuda 10.2. (For other CUDA version, please install the corresponding packages)

Data Preprocessing

Run the following commands to download the CNN/DM dataset, preprocess it, and save it locally.

mkdir data

• For ELIFE dataset

python preprocess.py --dataset elife

• For PLOS dataset

python preprocess.py --dataset elife

Train

Please run the scripts below:

bash run_train.sh

The trained model parameters and training logs are saved in outputs/train folder.

Inference

You can use our trained model weights to generate summaries for your data.

mkdir checkpoints
cd checkpoints

Step 2. Generate summaries for CNN/DM Test set.

bash run_inference.sh

After running the script, you will get the results in outputs/inference folder including the predicted summaries in generated_predictions.txt and the ROUGE results in predict_results.json.

Inference

• We thank the authors of Season their open-source codes.