Knowledge graph embedding and completion based on entity community and local importance

Knowledge graph completion can solve the common problems of missing and incomplete knowledge in the process of building knowledge graphs by predicting the missing entity and relationship information in the knowledge base. To the best of our knowledge, existing knowledge graph completion algorithms seldom consider the influence of entity communities, and no algorithm further considers the influence of local importance based on entity communities. In this paper, we propose a knowledge graph embedding model and completion method based on entity feature information. First, we use the community detection method to divide the knowledge graph into different entity communities, and calculate the local importance of the entity in the community. Next, we apply community information to obtain entities and relationships with low similarities to construct more appropriate negative triples. A new hybrid objective function that can simultaneously reflect the importance of entities and the structure of the knowledge graph is proposed to obtain high-quality entity and relationship embedding vectors to complete the knowledge graph. On the FreeBase and WordNet datasets, through comparison with six well-known knowledge graph completion methods, the experimental results show that our proposed algorithm has good completion performance.