Research And Application Of Representation Learning Algorithm For Knowledge Graph

With the rapid development of the times,we have gradually entered an era of informatization and intelligence.For the problem of large-scale knowledge graph data sparseness,many researchers embed the knowledge graph into a continuous vector space to solve it.In recent years,the representation learning technology represented by deep learning has set off a wave in the field of artificial intelligence,and has attracted much attention in many fields such as speech recognition,image analysis and natural language processing.Research on knowledge graph-based representation learning implements distributed representation of entities and relationships,significantly improves computational efficiency,effectively alleviates data sparse problems and can achieve fusion of heterogeneous information,and obtains effective information from massive data Has great significance.In this thesis,based on the analysis of knowledge graph-oriented representation learning related principles and technologies,combined with generative adversarial networks and word vector representation technology,this thesis proposes its own model and builds a question and answer system based on knowledge graph.The specific research content and results includes the following several aspects:(1)In view of the problem that the negative samples generated by the existing Trans series algorithms are not good enough,inspired by the generation of confrontation networks,the KBGAN model is proposed.Based on the model,TransE and TransD are used as discriminators,and DISTMULT and COMPLEX are used as generators.A total of four different combinations of experiments have been conducted.The results show that the performance of the proposed KBGAN representation learning algorithm is better than the existing Mainstream algorithms.And the KBGAN model has wide applicability and is not constrained by external ontology.(2)A representation learning algorithm based on TEKE is proposed.The existing representation learning algorithm mainly expands its structure on the basis of TransE to achieve the purpose of improving the performance of the algorithm,ignoring the rich text information in the knowledge base.This thesis builds a co-occurrence network of text corpus based on entity annotations,and on the basis of this co-occurrence network,connects entities and words together,and uses rich text information to performrepresentation learning on the knowledge graph.Through a large number of comparative experiments on the standard data set,it is verified that the model we proposed has achieved better results than the original model in most cases,solved the problem of complex relationship modeling,and to a certain extent Alleviates the impact of knowledge graph data sparseness on the performance of the representation learning model.(3)Using crawlers to collect data from web pages,construct knowledge base triple structure information,and design a question and answer system based on knowledge graph.Obtain web page data through the pyspider crawler framework,store it in the MongoDB database,convert the data into a triplet form,realize the question entity recognition of the question understanding part of the question and answer system through the LSTM + CRF model,get the target attribute through the keyword collection Chapters 3 and 4 show that the learning algorithm implements the answer acquisition part of the question and answer system.The background processing part and front-end display module of the system are completed through the Python flask framework,and the question and answer system page is displayed.