nayoung-oh / easy_to_read

CS376 Team 4 Repository about generating Easy-to-Read sentences considering user's fluency

Easy-to-Read Sentence Generation for the Intellectual Disability

CS376 Team 4 Code Repository (GITHUB)

Prepare wikilarge or wikismall dataset

1. Download the raw datasaet from DRESS repository (DRESS_REPO)
2. Unzip the file to wikilarge or wikismall folder
3. Run python preprocess_data.py --data wikilarge for wikilarge, python preprocess_data.py --data wikismall for wikismall

Set up environment

1. Run conda create --name --file requirements.txt python==3.7
2. If you want to run the demo website, install Flask, too.
3. If you want to generate EASSE reports, follow the EASSE repository (EASSE_REPO)

How to train the baseline model (naive transformer)

ex) with wikilarge dataset

python train.py --data wikilarge --model naive --loss none

How to train our own model (feature-based transformer)

ex) with wikilarge dataset, without weighted loss

python train.py --data wikilarge --model feature --loss none

ex) with wikilarge dataset, with weighted loss

python train.py --data wikilarge --model feature --loss weighted

How to test the trained model (calculate cosine similarity)

ex) with wikilarge dataset, with weighted loss

python test.py --data wikilarge --model feature --loss weighted --path PATH

if you want to check the output of a specific sentence, use simplify function

If you have any questions, use Issues