Mining Top-k Summarization Patterns For Knowledge Graphs

In recent years,under the support of network information technology,knowledge graphs containing a large number of unstructured and heterogeneous data represented by Wikipedia,Yago,Freebase,etc,have been rapidly developed.However,the data of knowledge graph has the characteristics of large volume,rich content,various types and lack of unified pattern description.Therefore,extracting knowledge graph patterns information and forming a digest pattern has important research significance for improving knowledge retrieval and mining quality.The graph patterns directly based on the existing pattern mining algorithm have the following problems:1)the efficiency of mining is low;2)it is difficult for users to control the frequent value of the algorithm,and often generates a large number of frequent subgraphs;3)Different patterns often overlap with each other redundantly.At the same time,the existing research results cannot solve the above problems well.Aiming at these shortcomings,this paper proposes a new summarization pattern mining method,which models the summarization pattern mining problem as an optimization problem,and proves that the objective function satisfies the sub-modulus and uses the mathematical properties of the marginal benefit maximization of the quadratic function.Solve the Top-k summarization patterns set.The experimental results show that the summary pattern mining method proposed in this paper is superior to the existing bi-objective function model method in mining quality and mining efficiency.The research work and innovations of this paper are as follows:The definition of the taxonomy summary patterns.First of all,the definition of the summary is given,which is the summary that satisfies the graph simulation matching condition.Secondly,considering that the constraints of the node labels in the graph matching need to be completely consistent,which will affect the practical application of the summary,this paper proposes the definition of the summary considering the taxonomy structure of labels based on the definition of the summary patterns.The quality of different summarization patterns is also different.Therefore,the quality measurement method of the summarization patterns is given.These theoretical research results have laid a solid theoretical foundation for the modeling and solving of summary patterns mining problems.Researches on summarization patterns mining problem modeling.This paper models the summarization patterns mining problem as a submodular function optimization problem.The objective function does not need to define parameters,and solves the problem by maximizing the marginal benefit of the submodular function.Finally,the validity of the objective function is verified in the real datasets.