Knowledge Graph Completion Model Based On Importance Priority

Knowledge graph technology is one of the current popular research directions,which can powerfully improve the efficiency of intelligent applications and therefore has received a lot of attention from researchers.In recent years,many research works for knowledge graph completion have been proposed in order to overcome the problem of data sparsity of knowledge graphs.However,most existing knowledge graph completion models ignore the features of triplet importance when learning knowledge,resulting in models that do not distinguish the importance of different information when learning triplet information.In this paper,we propose a novel knowledge graph completion model based on importance priority(Translating Embedding model based on Page Rank,PRTrans E),which considers the importance features in triplets and is divided into three main modules: entity importance module,relation importance module and triplet importance module.In the entity importance module,the PRTrans E model uses the Page Rank method to estimate the importance of entity nodes.In the relation importance module,this paper proposes a multi-grained relation importance estimation(MG-RIE),which provides a reasonable estimation of the overall importance of a relation by considering both the first-order importance and the higher-order importance of the relation.In the triplet importance module,the PRTrans E model takes different treatments for positive and negative triplets respectively to obtain the corresponding triplet importance scores.Further,this paper improves on the PRTrans E model and proposes the Imp Trans E model(Translating Embedding model based on Triplet Importance,Imp Trans E).In the entity importance module,the Imp Trans E model proposes the KGNode Rank method based on Page Rank.The KGNode Rank method estimates the importance ranking of entity nodes by considering both the importance of associated nodes and the probability of importance transfer.In the relation importance module,the Imp Trans E model also uses the multi-grained relation importance estimation method(MG-RIE).In the triplet importance module,the Imp Trans E model combines the importance of head and tail entities of the triplet and the relation importance as the importance score of the triplet.Through the above approach,the PRTrans E model and Imp Trans E model can assign different levels of attention to different triplet information during the learning process,which improves the representation learning performance of the models and thus achieves better completion effect.The experimental results show that both PRTrans E model and Imp Trans E model obtain better completion performance compared with the five comparison models in the two types of knowledge graph datasets.