PyTorch implementation of [CIKM 2021] Match-Ignition: Plugging PageRank into Transformer for Long-form Text Matching.
pip install -r requirements.txt
cd transformers-v4.30.0.dev0
pip install -e .cd data/dataset/cnse
tar xzvf orig.tar.gzNote: the original dataset can be downloaded from here.
# CNSE dataset
python generate_data.py --data_dir=data/dataset/cnse/orig --save_dir==data/dataset/cnse/model --from_raw_text=0 --append_keyword=1
# general raw-text dataset
python generate_data.py --data_dir=data/dataset/yuqing_news/example/orig --save_dir=data/dataset/yuqing_news/example/model --from_raw_text=1 --append_keyword=0python run.py If you use Match-Ignition in your research, please use the following BibTex entry.
@inproceedings{pang2021matchignition,
title={Match-Ignition: Plugging PageRank into Transformer for Long-form Text Matching},
author={Liang Pang and Yanyan Lan and Xueqi Cheng},
booktitle = {Proceedings of the 30th ACM International Conference on Information and Knowledge Management},
series = {CIKM'21},
year = {2021},
}
Copyright (c) 2019-present, Liang Pang (pl8787)