Research On Knowledge Graph Completion Method Based On Entity Semantics And Adjacency Information

The knowledge graph adopts a unified and structured approach to represent the rich semantics of the real world,which can efficiently manage and analyze factual knowledge.At present,knowledge graphs play an important role in tasks such as information retrieval,intelligent question answering,and information extraction.Although knowledge graphs contain a large number of triples,there are still incomplete situations.How to mine and supplement missing factual knowledge in existing knowledge graphs has received a lot of attention from researchers.In recent years,in the research of knowledge map completion methods,knowledge representation learning methods are mainly used to complete the research.In the process of knowledge representation,the semantic representation of fact knowledge will be inaccurate.According to the above problems,this paper has carried out research from two aspects of knowledge representation learning: the acquisition of real semantic information and the fusion of adjacent information.First of all,in order to solve the problem that the current knowledge map representation learning model does not accurately represent the entity semantics in the knowledge map,a learning semantics knowledge graph embeddings(LSKE)based on entity Semantic information modeling is proposed.Explicit modeling using planar encoding is used to represent entity semantics through the combination of horizontal and vertical vectors.Introduce semantic mapping matrix to capture entity specific semantics within the current training triplet;By utilizing the similarity between knowledge graph relationships,a shared semantic matrix is constructed,and different entities operate with the shared semantic matrix to preserve the shared semantics between entities.Secondly,in response to the problem of neglecting the adjacency information of entities in triplets during the training process of representation learning,which may result in incomplete semantic representation of sparse entities,a Learning context semantics knowledge graph embeddings(LCSKE)model integrating entity adjacency information is proposed,as well as the problem of generating negative example triplets that are useless for the training process during negative sampling,A neighborhood negative sampling(NNS)method based on the LCSKE model was proposed.The LCSKE model is based on the model LSKE and adds an auxiliary means of fusing entity adjacency information.It uses entity sparsity to determine the number of adjacent entities,and constructs adjacency information using entities and relationships.The entity adjacency information fusion is completed through a gate mechanism.The adjacency information negative sampling method utilizes adjacency information to prioritize the construction of negative example triplets that are helpful to the model during the negative sampling stage.Finally,link prediction and triplet classification experiments were conducted on publicly available datasets such as FB15k-237.Using corresponding evaluation indicators,conduct comparative experiments with existing knowledge representation learning models to verify the effectiveness of LSKE and LCSKE models and algorithms.