Relation Extraction Based on Relation Label Constraints

EasyChair Preprint no. 4604

Abstract

Knowledge graphs have a significant role in promoting natural language processing tasks, and they have received substantial attention. The relation extractor is a key step in the construction of a knowledge graph, so it is important to improve its performance. However, previous works are mainly based on the pipeline method, which rarely address the problem of  overlapping triplets. In addition, the literature does not consider models in which the correlation between relation pairs is addressed, which limits their accuracy. In this paper, we propose a new model called Relation Extraction Based On Relation Label Constraints(RRC) that is based on relation matrix constraints. The subject is extracted in our model in the first step; then, the relation and object are extracted based on the subject information. Each relation is regarded as a vector to assist in the extraction of the relation and object; the vector is used to consider the correlation between the relation vectors. This is used as a constraint to optimize the relation vector. Experiments on two public datasets, NYT and WebNLG, show that this method can perform well.

Keyphrases: BERT, Knowledge Graphs, Relation Extractor, relation matrix constraints

@Booklet{EasyChair:4604,
  author = {Kehua Miao and Kaihong Lin and Wenxing Hong and Chaoyi Yuan},
  title = {Relation Extraction Based on Relation Label Constraints},
  howpublished = {EasyChair Preprint no. 4604},

  year = {EasyChair, 2020}}