| model | offline score | online score | note |
|---|---|---|---|
| hmm | 45.80 | ||
| crf | |||
| bilstm+crf | 68.05 | ||
| bert | 78.14 | ||
| bert+crf | 77.18 | ||
| bert+span | 79.31 |
python main.py -m=bilstm_crf -b=200 -e=32 -mode=2 python main2.py -m=bert -b=200 -e=32 -mode=2 python main2.py -m=bert_crf -b=200 -e=32 -mode=2 python main3.py -m=bert_span -b=200 -e=32 -mode=2
python main2.py -m=bert -o=predict python main3.py -m=bert_span -o=predict -b=1
Tesla P100
16G
cuda9
python:3.6
torch:1.2.0.dev20190722
hmm模型
bilstn+crf
bert, bert+crf
bert+span
BIO B: 命名实体的起始 或 单个字命名实体 I: 命名实体的中间位置 或 结束位置 O：非命名实体
BIOES B: 命名实体的起始标注 I: 命名实体的中间标注 E: 命名实体的结尾标注 O: 非命名实体 S: 单个字命名实体
[1] CLUENER2020
[2] luopeixiang/named_entity_recognition
[1] Bidirectional LSTM-CRF Models for Sequence Tagging
[2] Neural Architectures for Named Entity Recognition
[3] BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding