Research On Entity Relation Extraction Technology Based On Deep Learning

Natural language processing technology is widely used,and building domain knowledge graph is one of its main application directions.As the underlying knowledge storage structure,the knowledge graph can provide support for upper-level applications such as intelligent Q & A,news recommendation,and information retrieval.The essence of the knowledge graph is a semantic network with a large number of entities as vertices and edges as relations.Therefore,entity relationship extraction technology is the basis and core of determining the confidence and availability of the knowledge graph.Entity relationship extraction technology can analyze natural language texts and identify corresponding entity relationships.In today's era of network data explosion,it can greatly save people's time to obtain information.Therefore,it is necessary to study entity relationship extraction technology for knowledge graph construction.This work takes the construction of domain knowledge graph as a starting point to study Chinese entity relationship extraction technology.Due to the lack of high-quality and largescale Chinese data sets in the military field,this thesis adopts the entity relationship extraction method based on remote supervision.The main work of this article includes:(1)In order to solve the problem of entity recognition of domain feature vocabulary and complex nested vocabulary,this thesis designs an entity recognition model based on word joint embedding,provides domain vocabulary feature learning for the model by constructing a domain vocabulary,uses BERT+Bi LSTM+CRF model architecture and designs BIO coding format for complex vocabulary,so as to realize the entity recognition model for specific domain.(2)The relationship extraction method based on remote supervision requires the domain knowledge base to support automatic labeling.Therefore,this thesis first builds a domainspecific knowledge base,and then realizes automatic labeling of large-scale texts through remote supervision and designs a relation extraction model based on a piecewise convolutional neural network model.Aiming at the problem of incorrect labeling in remote supervision,a sentence-level attention mechanism and a relationship-level attention mechanism are incorporated into the model to reduce the impact of incorrect labels.Through experimental analysis,the entity relationship extraction model used in this article has shown good results in both entity recognition and relationship extraction.(3)Based on the well-built complex entity recognition model and relationship extraction model,a knowledge graph-oriented entity relationship extraction system is designed,which realizes the functions of automatic crawling,text processing,and relationship extraction for specific domain data.Neo4 j graph database is used to store the corresponding entity relationship to construct Chinese domain knowledge graph.This system provides a visual display on the web side.Through system testing,the entity relationship extraction system oriented to knowledge graph developed in this article has good performance.Through research on entity recognition and relationship extraction,this article realizes the entity relationship extraction system for specific domain knowledge graph construction,solves the problem of poor expression of the existing entity recognition model and relation extraction model facing the domain text,and provides strong support for the construction of domain knowledge graph.