Domain Knowledge Graph Completion Based on Attribute Hierarchy

Knowledge graph completion is to solve the problem of lack of entities and relations in knowledge graphs. Existing Knowledge graph completion methods mainly embed entities and relations into latent vectors, and numerous researches have taken the rich relationships in knowledge graph into account such as category, entity, relation and semantic. However, only a few studies consider the relation between attributes, which is the basis of describing the entity. This paper proposes the Attribute Hierarchy Knowledge Graph Completion (AH-KGC) method, aiming at leveraging the attribute relation to find the missing obligatory property of entities. Primarily, in AH-KGC, we have discussed the attributes prerequisite relation, which can be described as a tree-like hierarchical structure, and then adopt the search algorithm of preorder traversal based on the hierarchy to find out the missing attributes. Specifically, we prove that attribute prerequisite is a special case of implication, thus can obtain attribute hierarchy from implications relation, which can easily be obtained in much mature research such as expert systems and make full use of the knowledge in various fields to make up for the vacancy of domain knowledge in the knowledge graph. The experiment has been performed on the CN-DBpedia and Probase datasets. The result demonstrates that AH-KGC can effectively complete the missing attributes of entities in knowledge graphs and achieve 100% accuracy under our evaluation system.