Chinese Named Entity Recognition

model: Bert+BiLstm
loss function: Cross Entropy Error Function

Chinese NER datasets:

• chinese weibo dataset
  
• chinese ontonote4
  
• chinese Resume
  
• chinese MSRA
  

how to train

(1) download chinese bert model

• git clone https://huggingface.co/bert-base-chinese

(2) install transformer

• pip install transformers
• conda install -c huggingface transformers

(3) change your own data file path

• form file config/config.py change bert_data_path, bert_tag_path, bert_model_path, bert_vocab_path

(4) learning rate option

• [2e-5, 3e-5, 4e-5, 5e-5]

result

datasets	P	R	F1
weibo	0.6866	0.6932	0.6899
Resume	0.9214	0.9497	0.9353
MSRA	0.9616	0.9104	0.9353
ontonote4	0.8344	0.7404	0.7846

reference

• https://github.com/huggingface/transformers
• https://arxiv.org/abs/1810.04805
• https://ieeexplore.ieee.org/document/818041