Development and Challenges of Translation-based Models for Knowledge Graph Completion

Abstract

Knowledge graphs are useful for many artificial intelligence tasks. However, knowledge graphs often have missing facts. To populate knowledge graphs, the graph embedding models map entities and relations in a knowledge graph to a vector space and predict unknown triples by scoring candidates triples. Translation-based models are part of knowledge graph embedding models and they employ the translation-based principle. The principle can efficiently capture the rules of a knowledge graph, however TransE, the original translation-based model, has some problems. To solve them many extensions of TransE have been proposed. In this paper, we discuss such problems and models.