domenicrosati / synscipass

SynSciPass: detecting appropriate uses of scientific text generation

SynSciPass: A dataset of synthetic scientific texts for detecting appropriate and inapproriate uses of text generation

This repo accompanies the paper (SynSciPass: detecting appropriate uses of scientific text generation)[] and contains the SynSciPass dataset (./data)

SynSciPass

The SynSciPass dataset is contained in ./data with the train, test, and validation splits at synscipass_.

Steps to Reproduce

Requires python 3.8 or higher.

$ pip install -r requirements.txt

Note: all identifying information is overwritten with anon.

Dataset construction

Run the notebooks in the SynSciPass creation folder.

Model training

Run the notebooks in the DAPT-TAPT and Train SynSciPass folders.

Experiments

Run the notebooks in the Experiment # folders.