Research On The Model Of Structure-augmented Knowledge Graph Embedding

In recent years,knowledge graphs have received considerable attention because of their ability to express rich information and their potential for use in Knowledgebased reasoning.Knowledge graph embeddings project the entities and relationns of knowledge graphs into vectors that are dense and low dimensional,thus allow-ing the complex semantic information and relations between these entities to be measured efficiently.However,traditional knowledge graph embedding methods consi-der only direct facts,making it difficult to achieve reasonable embedding learning of entities and relations when faced with sparse data.To settle this issue,this paper proposes a knowledge graph embedding repressentation model based on the combination of tensor decomposition and rule learning method.The model can iteratively perform embedding learning and rule learning to continuously generate new facts in the form of triples in order to achieve better performance in knowledge reasoning.First,we perform embedding learning on the primitive knowledge triples.Then,the rules are inferred,and new fact triples are generated based on these rules,and iteratively embedded into the model.In addition,this paper also proposes an iterative strategy to alternately implement knowledge graph embedding learning and rule learning to continuously generate new fact triples and obtain better knowledge graph embedding performance.First,we use the triples from a given data set as input and perform embedding learning to obtain the embedding of entities and relationships.Then,we infer and score the rules,generate new fact triples about sparse entities according to the rules with higher scores,and then iteratively embed the newly generated triples.In order to verify the performance of our proposed model,we conducted comparison experiments with the other five representative knowledge graph embedding models on the WN18 and FB15 k datasets.Experimental results show that our model has better performance in link prediction than the other models,and has significantly better performance than the other latest knowledge graph embedding models in the face of sparse data.