rillaha / CNN_LSTM_CRF_NER

neural network architecture to do named entity recognition

CNN_LSTM_CRF_NER

neural network architecture to do named entity recognition

1. 模型架构

底层CNN提取character embedding + 预训练的Glove word embedding
中间bi-LSTM
上层CRF进行标注

2. 框架

   theano + lasagne

3. NER标注

1）输入格式：参考data/test.txt，文档开头#begin doc_id num_of_sentences, 每个token一行，每句之间以空行隔开；

2) 测试结果：./test.sh

3) 结果处理：结果处理相关函数都在divide_text.py文件中

4. 训练

   1）文本格式与3中相同    2）参照run2.sh