/LET

Source code for paper "LET: Linguistic Knowledge Enhanced Graph Transformer for Chinese Short Text Matching", AAAI2021.

Primary LanguagePythonMIT LicenseMIT

LET

Source code of AAAI2021 paper "LET: Linguistic Knowledge Enhanced Graph Transformer for Chinese Short Text Matching".

Requirements

  • python: 3.7.5
  • mxnet-cu100: 1.5.1.post0
  • gluonnlp: 0.8.0
  • jieba: 0.39
  • thulac: 0.2.1
  • pkuseg: 0.0.22

Training

Before training, please contact the author of BQ and LCQMC dataset to download them. Then, you need to process the data to get the same format as the file data/LCQMC/train.json.

$ python utils/preprocess.py -i data/LCQMC/train.txt -o data/LCQMC/train.json

Train the model:

$ python train_sememe.py -c config/train_sememe_LCQMC.conf

The models trained by us can be downloaded from LET_BQ (password: aif3) and LET_LCQMC (password: udbv).

Cite

If you find our code is useful, please cite:

@inproceedings{lyu2021let,
  title={LET: Linguistic Knowledge Enhanced Graph Transformer for Chinese Short Text Matching},
  author={Lyu, Boer and Chen, Lu and Zhu, Su and Yu, Kai},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={35},
  number={15},
  pages={13498--13506},
  year={2021}
}