This repositary contains the code and data of "MultiOpEe:A Corpus of Multi-Perspective News Editorials." in NAACL'21.

Our dataset follows the below structure

The below two graphs show the statistics of our dataset and distribution of topics that our dataset covers.

transformers==4.5.0
torch==1.8.1

To reproduce the result, download the stance and relevance classifier from this google drive, modify their path in eval.sh, and simply run sh eval.sh

This should reproduce exactly the same result as we show in the paper.

Our best trained model is also available in this google drive. It is a multi-task BART-based model that uses both relevance and stance classification tasks as auxiliary signals.

Refer to train.py and train_both_auxiliary.py for training from scratch. More instructions of usage will be added soon.

@inproceedings{LCUR21,
    author = {Siyi Liu and Sihao Chen and Xander Uyttendaele and Dan Roth},
    title = {{MultiOpEd: A Corpus of Multi-Perspective News Editorials}},
    booktitle = {Proc. of the Annual Conference of the North American Chapter of the Association for Computational Linguistics (NAACL)},
    year = {2021}
}