Research On Knowledge Graph Embedding Method Based On Relational Context And Bidirectional Rotation Of Relationship

Knowledge graph has attracted extensive attention from industry and academia in the field of Big data and artificial intelligence,and has derived many intelligent applications based on Knowledge graph,such as recommendation system,intelligent question answering,semantic search,etc.However,due to the incompleteness of the Knowledge graph,these applications are limited.In order to complete the Knowledge graph automatically,the completion method based on Knowledge graph embedding has been widely studied.Therefore,how to efficiently and accurately complete the Knowledge graph has become the primary task of the embedded model.The performance of the knowledge graph embedding model is largely reflected in its ability to model relational patterns and mapping relationships.However,in the existing knowledge graph embedding learning,the structural information of the knowledge graph is underutilized,the semantic information of the triad is difficult to learn,and the embedding model does not have sufficient relational modeling capability.In order to solve the above pronlems,the main works in this paper are as follows:(1)To address the problem that most current embedding methods fail to make full use of triadic structure information,ignore the semantic information of relations as linked head and tail entities,a knowledge graph embedding model Cont E based on relational context is proposed.Cont E can model and infer different relationship patterns by considering the context of the relationship,which is implicit in the local neighborhood of the relationship.The forward and backward impacts of the relationship in Cont E are mapped to two different embedding vectors,which represent the contextual information of the relationship.Then,according to the position of the entity,the entity`s polysemous representation is obtained by adding its static embedding vector to the corresponding context vector of the relationship.Cont E is a fully expressive model.Experimental results on UMLS,Nations,FB15k-237 and Countries demonstrate that Cont E outperforms existing state-of-the-art baselines for the missing link prediction task.(2)To address the problem that most current embedding models cannot model complex mapping relationships and relational patterns simultaneously,we propose ECBR,a knowledge graph embedding model based on entity contexts and bidirectional rotation of relationships.The method first constructs the embedding vectors of entities and relations based on the complex space,then introduces the context information of entities to obtain the updated entity embedding vectors,and then maps the entities to the complex plane space.Then,it introduces the bidirectional rotation of relations to map the relations into forward rotation relations from entities to tail entities and reverse rotation relations from tail entities to head entities;then combines the entity embedding and the bidirectional rotation of relations to obtain the final embedding model.The ECBR model has a complete relational modeling capability,not only reasoning about the four relational models,but also modeling complex mapping relationships.In inclusion,link prediction comparison experiments were conducted on the benchmark datasets FB15k-237,WN18 RR and FB15 K.The experimental results revealed that the model improved in all metrics,which verified the effectiveness of the model.