luxinyu1 / Neural-Readability

Neural approaches to text readability.

Neural-Readability

Implementation of representative neural (supervised and unsupervised) approaches to measuring readability. Currently, this repo only focuses on sentence-level readability.

Unfortunately, Newsela is not a publicly available dataset. Only place the raw Newsela file in ./datasets/newsela will make this repo works.

Usage

• Finetune a BERT model to do readability classification task:

   python ./neural_readability/finetune.py

• Train a BiLSTM model to do readability classification task:

   python ./neural_readability/train.py

Unless otherwise specified, the training / validation / testing logs should be found in ./logs/. More usage scripts can be found in ./scripts/.

Trained models

Arch	dataset	Link
BERT	Newsela	Download
BiLSTM	Newsela	Download

Acknowledgements

Some code in this repo is based on GRANT. Thank for its wonderful works.