athoszhou/crf

pytorch 条件随机场实现序列标注

crf

pytorch 条件随机场实现序列标注

模型下载

bert-vocab-chinese
bert-model-chinese
Traditional-Chinese ELMo

Pre-requirements

• must python >= 3.6 (if you use python3.5, you will encounter this issue HIT-SCIR/ELMoForManyLangs#8)
• pytorch 1.0
• opencc
• elmoformanylangs
• pytorch_pretrained_bert

简体转繁体

# 安装
# pip install opencc-python-reimplemented

# t2s - 繁体转简体（Traditional Chinese to Simplified Chinese）
# s2t - 简体转繁体（Simplified Chinese to Traditional Chinese）
# mix2t - 混合转繁体（Mixed to Traditional Chinese）
# mix2s - 混合转简体（Mixed to Simplified Chinese）

import opencc
cc = opencc.OpenCC('s2t')
s = cc.convert('你好，吃饭了吗？')
print(s)

Result

Network + CRF	acc	precision	recall	f1-score
Bi-GRU	0.9638	0.9294	0.9264	0.9279
2 layer Bi-GRU	0.9613	0.9404	0.9362	0.9383
3 layer Bi-GRU	0.9644	0.9366	0.9433	0.94
bert	0.9655	0.9531	0.9556	0.9543
bert Bi-GRU	0.9648	0.9572	0.9511	0.9541
elmo Bi-GRU	0.9649	0.9468	0.9394	0.9431