Entity and relationship joint extraction method based on head entity prediction

Abstract

The invention discloses an entity and relationship joint extraction method based on head entity prediction. According to the method, an entity and relationship extraction task is decomposed into two sequence annotation tasks, namely head entity identification annotation and tail entity identification annotation, by only integrating a relationship into an annotation strategy of a tail entity label;the vector of a head entity is used as the input of a tail entity annotation identification task, and the effect of the model is improved by using the thought of a prior probability. The annotated label of the tail entity determines the relationship between the entity pairs while completing tail entity identification, so that the problem of incomplete extraction in the presence of an entity overlapping phenomenon and the problem that extraction cannot be performed by a traditional method in the presence of entity nesting are solved.

Description

[0001] 1. Technical field [0002] The invention relates to the field of natural language processing, in particular to information extraction, in particular to a joint extraction method of entities and relations based on head entity prediction. [0003] 2. Background technology [0004] Today is the era of knowledge economy. With the vigorous development of Internet technology and the continuous improvement of social informatization, data resources are growing explosively, followed by the accumulation of massive text data. How to quickly and accurately obtain the required information from a large number of unstructured text data resources has attracted more and more people's attention, and entity and relationship extraction is a technical means that emerged as the times require. Entity and relationship extraction refers to the extraction of entity pairs and the relationship between entity pairs from unstructured text. An entity pair refers to two entities that may have a relati...

AI Technical Summary

Problems solved by technology

[0010] The purpose of the present invention is to solve the problem of incomplete extraction of triples when there is entity overlap in the corpus, and the problem that traditional methods cannot extract when there is entity nesting in the corpus

The labeling label of the tail entity determines the relationship between entity pairs while completing the tail entity recognition, thus solving the problem of incomplete extraction when there is entity overlap and the problem that traditional methods cannot extract when there is entity nesting

Images