This is the source code for the paper ''Adversarial Transfer Learning for Chinese Named Entity Recognition with Self-Attention Mechanism'' accepted by EMNLP2018. The paper can be download from http://aclweb.org/anthology/D18-1017.
- TensorFlow >= v1.2.0
- numpy
- python 2.7
Please download the WeiboNER dataset, SighanNER dataset and MSR dataset, respectively. The dataset files are put in data directory.
For training the model on WeiboNER dataset, you need to type the following commands:
- python preprocess_weibo.py
- python train_weibo.py
For SighanNER dataset, the operation is similar.
We have provided our best model on the original WeiboNER dataset in the ckpt directory. You just run the model like:
- python preprocess_weibo.py
- python test_weibo.py
In addition, if you adjust certain hyper-parameters and train the model, you can test the model with restoring certain checkpoint.
If you use the code, please cite this paper:
Pengfei Cao, Yubo Chen, Kang Liu, Jun Zhao. Adversarial Transfer Learning for Chinese Named Entity Recognition with Self-Attention Mechanism. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP2018).