Research On Knowledge Graph Completion And Entity Alignment Method Based On Embedding Technology

The task of knowledge graph completion is to predict the missing relations between entities according to the existing knowledge triples in the knowledge graph to form new triples.The entity alignment task is to find the entities that refer to the same real-world entity in different knowledge graphs,so as to fuse multiple knowledge graphs.Knowledge graph embedding represents the entities and relations in a knowledge graph as low-dimensional vectors in a continuous vector space,which provides a good technical basis for the above two tasks.A knowledge graph embedding model should express as many different types of relation connectivity patterns and mapping properties as possible,so as to capture more abundant semantic information of entities and relations,and then more accurately predict the missing relations between entities.Embedding-based entity alignment methods,which perform entity alignment by measuring the similarity between entity embeddings,have achieved promising results.However,existing methods are still challenged by the error accumulation of embeddings along multi-step relation paths and the semantic information loss.In this paper,we respectively propose a novel knowledge graph embedding model(called Linea RE)and two entity alignment methods(called Vec Dist and RSim EA,respectively)for the above problems.Linea RE regards knowledge graph embedding as a simple linear regression task,which can well express four connectivity patterns(symmetry,antisymmetry,inversion and composition)and four mapping properties of relations(one-to-one,one-tomany,many-to-one and many-to-many).Moreover,we give a strict mathematical proof of linea RE's modeling abilities.Vec Dist aligns entities and aligns relations at the same time,and shortens the distance between the vectors of the aligned relations in a certain way,so as to reduce the problem of embedding error accumulation on the multi-step relations paths.RSim EA further introduces a method to calculate the structural similarity between relations and aligns relations with high structural similarity,effectively reducing the negative impact of semantic information loss.Experimental results on several datasets show that our proposed Linea RE,Vec Dist and RSim EA have achieved achieve state-of-the-art results on knowledge graph completion and entity alignment tasks,respectively.