EHP:Entity Hyperplane Projection for Knowledge Graph Embedding with Entity Descriptions

Knowledge graph embedding aims to project entities and relations into a low-dimensional and continuous vector space to represent the semantic information of entities and relations. Most existing models of knowledge graph embedding only concentrate on the structured information of the knowledge graph and merely use knowledge triples that can also be called the fact triples to learn the representation of entities and relations, but ignore some useful information that emerge in text descriptions of entities. To this end, this paper proposes the entity hyperplane projection (EHP) model. EHP learns both from knowledge triples and text descriptions, project the entity embedding onto the semantic hyperplane corresponding to entity descriptions to build interactions between the two information sources. Extensive experiments show that EHP model achieves substantial improvements against baselines on the tasks of knowledge graph completion and entity classification.