How to run the code:

python reranker.py train DEBUG B

where
train is training and save model first, if using other word, then just load models without training.
DEBUG use small dataset. If using other word or None, use standard data.
B means LSTM with CNN model.

Data format

Generate the nbest results using neural BI-LSTM-CRF at NbestNER.
Generate the nbest results using discrete CRF model at CRF++
Data format follow the sample data

Folder

data: where the data saved model: model file saved results: training model saved utils: load data, metric file saved

Cite:

@article{yang2017neural,  
 title={Neural Reranking for Named Entity Recognition},  
 author={Yang, Jie and Zhang, Yue and Dong, Fei},  
 booktitle = {Proceedings of RANLP},
 year={2017}  
}  

Version

2017-April-4: init version