Research On Knowledge Graph Reasoning Algorithm Based On Tensor Decomposition

In the information age,big data contains a lot of useful knowledge,which is worthy of deep exploration,mining,research and analysis.As the information exploded and the amount of data skyrocketed,knowledge graphs have become more and more important and widely used in vertical search,NLP,social networks,semantic web,intelligent question answering,bio-informatics and medicine.The knowledge graph inference algorithm is based on the existing knowledge in the knowledge graph,and the process of obtaining new knowledge through computational reasoning is a hot topic of the current knowledge graph.For a large number of knowledge graph inference problems,the single feature of the data is only considered in the existing tensor decomposition knowledge graph algorithms,and the information utilization of the knowledge graph is insufficient.Therefore,based on the tensor decomposition(RESCAL)algorithm,a knowledge graph inference algorithm(KLLS-RESCAL algorithm)that fuses relative entropy-squared difference is proposed in this paper,which considers the information features of the knowledge graphs and the numerical features comprehensively.In the algorithm,the fusion loss function is set through weight parameters based on relative entropy and squared difference,which performs tensor decomposition from two perspectives of information theory and numerical analysis,and the complexity of knowledge graph data is considered comprehensively.The corresponding theoretical analysis and experimental results show that the fusion algorithm proposed in this paper has higher prediction accuracy than other single tensor decomposition algorithms in the knowledge graph inference.In addition,most of the knowledge graph data is very large.To solve the problem of large-scale data operation rate,a knowledge graph optimization algorithm(KLLS*-RESCAL algorithm)incorporating standardized relative entropy-square difference is proposed.The factor matrix in the algorithm is normalized firstly,then the standardized fusion loss function is set based on relative entropy and squared difference,and the knowledge graph is optimized from two perspectives of information theory and numerical analysis.The corresponding theoretical analysis and experimental results show that,this algorithm enjoys higher prediction accuracy than other tensor decomposition single algorithms in knowledge graph inference,and the operation speed is better than the KLLS-RESCAL algorithm.Finally,the two fusion algorithms are compared in this paper.Theoretical analysis and experimental results show that there is no significant difference in the prediction results of the two fusion algorithms,but the operation speed of the KLLS *-RESCAL algorithm is superior to the KLLS-RESCAL algorithm.