文章目录
任务介绍
NER (Named Entity Recognition)即命名实体识别。顾名思义就是识别文本当中的实体信息。举个例子,
输入:张三现在在武汉市江夏区金融港
输出:B-PER,E-PER,O, O,O,B-CITY,I-CITY,E-CITY,B-DISTRICT,I-DISTRICT,E-DISTRICT,B-LOCATION,I-LOCATION,-E-LOCATION
其中,“张三”以PER,“武汉市”以CITY,“江夏区”以DISTRICT,“金融港”以LOCATION为实体类别分别挑了出来。
NER的过程,就是预测出其序列标注的过程。
备注:
BIOES标注方式中分别代表什么意思:
B,即Begin,表示开始
I,即Intermediate,表示中间
E,即End,表示结尾
S,即Single,表示单个字符
O,即Other,表示其他,用于标记无关字符
相关论文
目前的解决方案及对应的paper如下,其中bilstm-crf和bert-crf成为ner领域常用的baseline。
BiLSTM-CRF
paper:Neural Architectures for Named Entity Recognition
IDCNN-CRF
paper:Fast and Accurate Entity Recognition with Iterated Dilated Convolutions
BERT-CRF
ner-mrc
paper:A Unified MRC Framework for Named Entity Recognition
BERT-CASCADE-CRF
paper:A Novel Cascade Binary Tagging Framework for Relational Triple Extraction
FLAT
paper:FLAT: Chinese NER Using Flat-Lattice Transformer
global-pointer
paper:Global Pointer: Novel Efficient Span-based Approach for Named Entity Recognition
efficient-global-pointer
blog:Efficient GlobalPointer:少点参数,多点效果
bert-biaffine
paper:Named Entity Recognition as Dependency Parsing
评测结果 to be continued
github
bert4torch 里面的轮子已经非常丰富了,咱就直接拿来用就好了