Unified Named Entity Recognition as Word-Word Relation Classification


论文:Unified Named Entity Recognition as Word-Word Relation Classification

代码:ljynlp/W2NER

1 创新思路

1、提出了一种基于word-word关系分类的统一NER建模模型(flat,nested,discontinuous)

2、该架构通过对实体词之间的相邻关系进行有效建模,用Next-Neighboring-Word(NNW)和Tail-Head-Word-(THW-)关系来解决统一NER的内核瓶颈问题

2 嵌套NER和不连续NER的示意图

3 NER as Word-Word Relation Classification

word-word矩阵,表示token对之间的关系,非对称,表示第行个token和第列个token之间的关系

4 整合流程