Research On The Key Technology Of Knowledge Graph Learning And Reasoning Oriented To Web Data

An increasing number of Linking Open Data and user-generated contents are published on the Web since Semantic Web has been proposed.This causes Web data to show characteristics of structure looseness,fragmented content and complex semantics,which has not been able to meet people's needs of fast and accurate information positioning.Under the circumstance of exponential growth in the scale of structured,semi-structured and unstructured Web data,Knowledge Graph provides smart solutions for the efficient processing of information.The Knowledge Graph enriches and expands the ontology at the entity level so that it can be used to describe the entities and their semantic relations in the real world,to help establish better understanding of information and interactive operation between people and machines and provides an important data support for the artificial intelligence.However,In addition to the explosion of Web data scale,the rapidly dynamic Web data also brings severe challenges to the updating of the Knowledge Graph.For example,in the process of scale expansion of the Knowledge Graph,structured knowledge is organized in the form of network only through its explicit relationship,which causes the latent relations between the Knowledge Graph entities not to be fully mined.Furthermore,the updating speed of Knowledge Graph obviously lags behind the changing speed of Web data,with a result that accuracy and timeliness of the semantic search is affected.Based on learning and analyzing the relevant theories and techniques of Knowledge Graph complement,this thesis carries out a research from the perspective of Knowledge Representation Learning and knowledge reasoning in order to optimize the performance of semantic search system and provide fast,accurate and efficient query services.Firstly,a Knowledge Representation Learning algorithm based on the semantic tensor of Knowledge Graph is proposed.Taking “Rescal tensor decomposition” as the core idea,combining the Knowledge Graph sampling technique and principal component analysis technique,and predicting the relations among knowledge entities by the methods of matrix learning,vector mapping and dimensionality reduce,the prediction efficiency of multivariate relationships is improved.Secondly,a method of Knowledge Graph completion based on Bayesian Reasoning is proposed.This method uses Bayesian probabilistic reasoning theory and RDF implication reasoning rules,to make the co-inference on the potential relations among the entities nodes,and the relations between new node and original node are predicted.The prediction accuracy of unknown relations and the efficiency of mining model potential factors are improved.Thirdly,a Knowledge Graph construction scheme based on domain ontology is proposed,a Knowledge Graph of library information is constructed through seven steps,some of these steps are knowledge collection,the establishment of entities relation etc.Finally,with a strong foundation of theoretical research,a prototype system is designed and implemented based on Knowledge Graph learning and reasoning techniques,information querying and related information pushing are deployed on it.Based on the above researches,this thesis proposed a more efficient and accurate Knowledge Graph learning and reasoning method in order to eliminate the affection of multi-source heterogeneous and highly dynamic Web data brought to the constructing,updating and completing of Knowledge Graph to provide better services on knowledge retrieval,to optimize querying process and enhance user's experience.With the development of Knowledge Graph technology,the Knowledge Graph learning and reasoning technology can be combined with Machine Depth Learning,Cloud Computing,Block chain,Large Data,and Biology gene engineering and other new fields to play a valuable role in society.