Joint extraction of Chinese entitys and relations based on hierarchical labeling

Entitys and relations triples are the basic information units that make up the knowledge graph, and represent some simple and specific information. This paper proposes a hierarchical annotation model based on BERT to jointly extract entity relation triples from text. First, the subject of the triples is marked by the labeling method. On this basis, for each predefined relationship, the half-pointer and half-marking method is used, and the double pointer is used to mark the beginning and end positions of the corresponding object in the text, and then the Complete extraction of triples can effectively solve the problem that traditional methods are difficult to extract triples with overlapping entities.