项目作者: Jackma3

项目描述 :
chinese ner(model: bert+lstm)
高级语言: Python
项目地址: git://github.com/Jackma3/NER-Bert-BiLstm.git
创建时间: 2021-03-09T10:07:47Z
项目社区:https://github.com/Jackma3/NER-Bert-BiLstm

开源协议:

下载


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
(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