The code for the SelfExplain framework (https://arxiv.org/abs/2103.12279)

Currently, this repo supports SelfExplain-XLNet version for SST-2 dataset. The other datasets and models shown in the paper will be updated soon.

Data for preprocessing available in data/ folder

On a python shell, do the following for installing the parser

>>> import benepar
>>> benepar.download('benepar_en3')

sh scripts/run_preprocessing.sh

sh scripts/run_self_explain.sh

(In Progress)

 python model/infer_model.py
        --ckpt $PATH_TO_BEST_DEV_CHECKPOINT \
        --concept_map $DATA_FOLDER/concept_idx.json \ 
        --batch_size $BS \
        --paths_output_loc $PATH_TO_OUTPUT_PREDS \
        --dev_file $PATH_TO_DEV_FILE

Coming Soon ...

@misc{rajagopal2021selfexplain,
      title={SelfExplain: A Self-Explaining Architecture for Neural Text Classifiers}, 
      author={Dheeraj Rajagopal and Vidhisha Balachandran and Eduard Hovy and Yulia Tsvetkov},
      year={2021},
      eprint={2103.12279},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}