Knowledge graph completion method and system based on multi-scale dispersed dynamic routing

Abstract

The invention provides a knowledge graph completion method and system based on multi-scale dispersion dynamic routing, and the method comprises the steps: enabling an obtained to-be-completed triple to circularly interact with a memory matrix through employing a multi-head attention mechanism, coding a potential dependence relation between an entity and a relation, and generating a triple coding vector; the triple coding vectors are input into the trained capsule network, global features are extracted, different coupling coefficients are distributed to the global features, missing triads are predicted according to the global features, and completion of the knowledge graph is completed; according to the method, the potential dependency relationship between the entities is effectively modeled during knowledge graph prediction, the deviation caused by dynamic routing during prediction is reduced as much as possible, and the characteristics of different abstract levels are captured, so that the completion prediction precision of the knowledge graph and the effect of triple classification are improved.

Description

technical field [0001] The invention belongs to the technical field of knowledge graphs, and in particular relates to a knowledge graph completion method and system based on multi-scale decentralized dynamic routing. Background technique [0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art. [0003] Knowledge Graph (KG) was proposed by Google in 2012 and is usually expressed as a multi-relational graph, which is a collection of three elements of facts. There are a large number of facts in the world that can be simply represented as entities and their relationships. YAGO, Freebase and DBpedia are usually KGs representing relationships between entities in the form of facts (head entity, relation, tail entity), represented as (h; r; t), e.g. ("Beijing", "belongs to", " China”); knowledge graphs can be used to describe ternary relationships in the fields of recommender systems,...

AI Technical Summary

Problems solved by technology

[0007] 1) MCapsE improves CapsE by making multi-scale convolution kernels in the convolutional layer, which improves the semantic features of the model to extract triples, but only expanding the capsule network model cannot effectively capture the potential dependencies between entities; 2) While R-MeN, although it can well extract the potential dependencies between entities in different semantic spaces, is limited by the disadvantages of low coding efficiency of convolutional neural networks; 3) and because the dynamic routing part in the capsule network uses The Softmax function will lead to wrong summation of the prediction vectors, which will affect the final result of the prediction. The DE-CapsNet model uses the decentralized dynamic routing with the Sigmoid function to improve the capsule network. In the field of image classification, the data sets CIFAR-10 and F -MNIST shows very good performance, but it has not been applied in the field of knowledge map completion

Images