Official Repository of KCD@NAACL'22

KCD: Knowledge Walks and Textual Cues Enhanced Political Perspective Detection in News Media
Wenqian Zhang*, Shangbin Feng*, Zilong Chen*, Zhenyu Lei, Jundong Li, Minnan Luo
In Proceedings of NAACL 2022, oral presentation

Links: [Paper link] [Oral Presentation Slides] [Wenqian Zhang homepage]

Details

1. main folder consists of codes for KCD model, semmain folder indeciates codes for Semeval dataset and allmain folder indeciates codes for Allsides dataset seperately.
2. sem/Train folder is Semeval training data.
3. if you need training data for allside dataset, you can click here
4. if you need Trained Model, you can click here

File structure:

├── main               # code for KCD
      ├── allmain      # code for KCD in Allsides dataset
            ├── KSD_Dataset.py
            ├── KSD_GatedRGCN.py
            ├── Tools.py
            └── Run_Model.py      # train model
      └── semmain      # code for KCD in Semeval dataset
            ├── KSD_Dataset.py
            ├── KSD_GatedRGCN.py
            ├── Tools.py
            └── Run_Model.py      # train model
├── sem      # data for Semeval dataset
└── all      # data for Allsides dataset, you need to download from google drive

Dependencies

Our code runs on the Titan X GPU with 12GB memory, with the following packages installed:

Python 3.8.5
torch 1.7.1
pytorch_lightning
numpy
torch_geometric
argparse
sklearn
pickle

How to reproduce

train the model by running

python Run_Model.py

Citation

If this paper inspires you, please cite us!

@inproceedings{Zhang2022KCDKW,
  title={KCD: Knowledge Walks and Textual Cues Enhanced Political Perspective Detection in News Media},
  author={Wenqian Zhang and Shangbin Feng and Zilong Chen and Zhenyu Lei and Jundong Li and Minnan Luo},
  booktitle={NAACL},
  year={2022}
}

Questions?

Feel free to open issues in this repository. You can also contact us at 2194510944@stu.xjtu.edu.cn.