Research On Knowledge Graph Completion Based On Triple Embedding And Random Path

The knowledge graph can structure and semantically store the objective facts of nature.These two characteristics have also made the knowledge graph attract more and more attention from academia and industry,and have become one of the basic technologies in artificial intelligence research.In the automatic construction of knowledge graphs,knowledge errors and lack inevitably occur,so knowledge graphs generally have the problem of data sparseness.To improve the integrity of the knowledge graph,many knowledge representation learning algorithms have emerged in recent years to reason new facts.This task is also called knowledge graph completion.However,existing research mainly focuses on the semantic and logical relationship between entity relations in knowledge graphs,often ignoring the structural relationship of triples in knowledge graphs.In addition,the scale of the knowledge graph continues to increase,and the judgment of the relationship between entities becomes more complicated.A series of problems such as long-tail problems and data sparsity have brought more challenges to knowledge graph completion.Traditional knowledge graph methods in link prediction tasks often limit the length of prediction paths.Today’s popular and widely used knowledge graphs often have large volumes and extremely high data volumes.This also causes the long-tail path problem,that is,the association between data has a long-term impact.The long tail path will exceed the limit of the predicted path length in the traditional method.Therefore,using traditional methods to obtain insufficient paths will result in missing some important nodes.The main work of this paper is as follows:Aiming at the insufficient acquisition of triplet features,a knowledge graph triple embedding model based on representation learning is proposed.The triple embedding model consists of three aspects: based on the principle of translation invariance,realize the calculation of triplet entity features;based on the principle of path resource allocation,path semantic feature calculation is realized;using the environmental structure assessment,the calculation of structural characteristics is realized.Finally,the acquisition of triple feature information is realized by calculating the triple feature of triple entity,relation and structure.Aiming at the long-tail problem and the problem of limited path jump length,this paper proposes a random path model.The random path no longer limits the jump length,and the limitation of path repetition is realized by creating a damping coefficient.By constructing a function to score path efficiency and node relevance,further optimization and screening of paths can be realized.Knowledge Graph Completion Based on Triple Embedding and Random Path.The random path model is used to screen high-efficiency sentences.Use the triple embedding model to obtain the triple features of a single node in a sentence.Bring them into the temporal neural network for semantic calculation and realize semantic transmission to remote nodes.The confidence score is proposed to detect the credibility of nodes,realize the judgment of knowledge graph completion nodes,and improve the quality of knowledge graph completion nodes.