Research And Implementation Of Open Information Extraction Method For Automatic Construction Of Knowledge Graph

Knowledge graph is a prevailing technology that widely used in fields such as finance,e-commerce,judicial system and so on.Compared with the traditional knowledge graph,the open knowledge graph studied in this paper does not need to define the type of relationship between entities in advance,so it has stronger domain adaptability and migration ability.In the Internet era,the explosive growth of text data on the Internet contains a large amount of high-value information,which provides a data basis for the automatic construction of knowledge graph.The goal of this paper is to study the method of extracting structured information from natural language text.It includes two parts:information extraction algorithm and extraction canonicalizing algorithm,so as to construct knowledge graph from public text data.Finally,based on the algorithms above,this paper designs and implements an open knowledge graph construction,query and display system.Specifically,this paper mainly includes the following three aspects of work:(1)Aiming at the defects in the representation of sentence grammatical features in the current research of information extraction,this paper proposes an open information extraction method based on deep learning.It uses the graph representation learning model TransD to learn the dependency representation between words,and then uses the learned dependency representation for open information extraction.Experimental results show that this method is superior to the current mainstream methods on three public Chinese and English data sets.(2)For the practical application,this paper proposes an open extraction canonicalizing method,which includes entity link algorithm based on attention mechanism and relationship canonicalizing method based on text similarity clustering.In the experiment,the entity link algorithm achieves great performance on three public Chinese and English data sets and real data,and the relationship canonicalizing method also has good performance on real data.(3)Based on the mentioned algorithms,this paper constructs an open knowledge graph and implements its query and display system.The system can automatically crawl the news text from the media website,extract the entities and relationships from the news,and use the extracted structured information to construct the knowledge graph.Users can search entities according to keywords,and the system shows an entity profile according to the basic information and relationship networks of the entity.