More Is Better: Sequential Combinations of Knowledge Graph Embedding Approaches

Abstract

Constructing and maintaining large-scale good quality knowledge graphs present many challenges. Knowledge graph completion has been regarded a promising direction in the knowledge graph community. The majority of current work for knowledge graph completion approaches do not take the schema of a target knowledge graph as input. As a result, the triples generated by these approaches are not necessarily consistent with the schema of the target knowledge graph. This paper proposes to improve the correctness of knowledge graph completion based on Schema Aware Triple Classification (SATC), which enables sequential combinations of knowledge graph embedding approaches. Extensive experiments show that our proposed approaches can significantly improve the correctness of the new triples produced by knowledge graph embedding methods.