domenicrosati / improving-lay-factuality-with-retreival

Improving Lay Summarization Factuality with Retreival

README

Improving the factuality of biomedical lay summarization with retreival. For the https://biolaysumm.org/ task.

Setup

Download and unzip the data file from https://codalab.lisn.upsaclay.fr/competitions/9541#participate-get_data

pip install -r requirements.txt
python -m nltk.downloader punkt
pip install https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.5.1/en_core_sci_sm-0.5.1.tar.gz

Data Characterization

See notebooks/characterization.ipynb

Enhance datasets

python main.py --dataset human_eval --task enhance --enhancement-method umls
python main.py --dataset human_eval --task enhance --enhancement-method wiki_keywords
python main.py --dataset human_eval --task enhance --enhancement-method wiki_lucene
python main.py --dataset human_eval --task enhance --enhancement-method scite

Summarization Baseline

python main.py --dataset elife --task finetune
python main.py --dataset plos --task finetune
python main.py --dataset both --task finetune

Evaluate Summarization

python main.py --dataset elife --task evaluate
python main.py --dataset plos --task evaluate
python main.py --dataset both --task evaluate