The Research And Application Of Entity And Relation Completion Based On Knowledge Graph

With the rapid development of the Internet technology,more and more data are stored on cloud servers and personal terminal devices,whether pictures,videos,audio or text data are unstructured,so they cannot be easily used by computers.Knowledge graph saves data in a structured form so that computers can effectively apply these data.At present,knowledge graphs have become an indispensable technology in search engines,question answering systems,knowledge reasoning and other fields.However,the construction and maintenance of knowledge graphs is a long process,which requires continuous improvement by experts in related fields.Therefore,knowledge graph completion has become a hot research topic in the field of knowledge graphs.It is dedicated to predicting missing entities and relationships,automatically improving the structure of knowledge graphs,and making them play a better role in downstream applications.It can reduce a lot of manual labeling work and is an important technology in the construction of knowledge graphs.This paper is based on the Elementary Mathematical Natural Language Understanding System,which works to understand unstructured mathematical questions into structured triplet representations.However,due to the limitations of design and implementation of the system,there is a lack of knowledge representation in this transformation process.Therefore,this paper proposes two knowledge graph completion methods for these problems,and the purpose is to improve the system's knowledge representation of mathematical problems.In summary,the work of this paper mainly includes the following aspects:(1)The problem of lack of knowledge representation in elementary mathematics natural language understanding system is analyzed.There are mainly two aspects of knowledge that cannot be extracted.The first is that the hidden entities not mentioned in the text title cannot be extracted.The second is that the correct relationships cannot be extracted due to the size of the labeled dataset and the high similarity threshold.(2)A rule-based entity auto-completion method is proposed,which mainly completes implicit entities.The writing standard of entity completion rules is defined,and a framework for entity completion is designed to implement operations such as parsing rules and generating triples.(3)A relationship auto-completion method based on deep learning is proposed,which mainly completes missing relationships.A deep learning model is designed and trained,in which external entity embedding vectors are introduced to enhance entity representation.The labeled triple samples are used for training,and the trained model is applied to the completion task of the mathematical knowledge graph.(4)A knowledge graph completion module that integrates the entity auto-completion method and the relationship auto-completion method is constructed,which is applied to the elementary mathematics natural language understanding system,and the accuracy rates of triples completed by the two completion methods are tested respectively.The correct rate of the entity auto-completion method is about 93%,and the correct rate of the relationship auto-completion method is about 75%.This module improves the problemsolving rate of the overall system by about 30%.