Font Size: a A A

The Research And Implementation On Joint Entity Relation Extraction Algorithm For Short Text

Posted on:2022-11-13Degree:MasterType:Thesis
Country:ChinaCandidate:W NongFull Text:PDF
GTID:2518306752953859Subject:Computer technology
Abstract/Summary:PDF Full Text Request
With the development of the Internet,there is more and more information on the Internet.Researches on how to dig out valuable information from massive amounts of information have a significant effect on the development of society.Obtaining structured knowledge from a large amount of unstructured data and constructing a knowledge graph can facilitate users to find the information that they need accurately and efficiently.The task of joint extraction of entity relations can solve this problem effectively,combining entity recognition and relation extraction tasks to one model which will reduce the error propagation loss between the two tasks.Also,it is useful and important to recognize entities and relations from the text to construct triples and build knowledge graphs.This paper analyzes the problems existing in the prior studies of joint extraction of entity and relation and proposes two models to improve the effect.Meanwhile,we build an extraction system to approve the practicability of our research.The main work of this paper includes:· Joint extraction based on hierarchical convolution of fusion graphA joint extraction model that uses fusion graph hierarchical convolution to en-hance feature representation is proposed.Syntactic information is integrated intothe model for the joint extraction of entities and relations,which solves the lack ofinteraction between the two tasks in the traditional pipeline model and the problemof unused dependency syntax information.Experiments on public datasets verifythe effectiveness of the model in this chapter.· Joint extraction based on dual encodersReferring to the sequence-to-sequence model in machine translation,this chapterproposes a joint extraction model for entities and relations based on dual encoders.Using graph convolution and Modify LSTM to encode sentences,and adopting afusion attention mechanism to decode.The model focuses on both the grammaticaland contextual features of the sentence.· The information extraction and visualization system of text Based on the proposed two joint extraction models,a prototype system for joint extraction of entities and relations is implemented.The user can extract the triples in the text through the visual interface,and modify the triples stored in the database.
Keywords/Search Tags:Information Extraction, Joint Extraction of Entity and Relation, Graph Neural Network, Sequence-to-sequence Model, Attention Mechanism
PDF Full Text Request
Related items