Knowledge graph embedding with concepts

Abstract

Knowledge graph embedding aims to embed the entities and relationships of a knowledge graph in low-dimensional vector spaces, which can be widely applied to many tasks. Existing models for knowledge graph embedding primarily concentrate on entity–relation–entitytriplets, or interact with the text corpus. However, triplets are less informative, and the in-domain text corpus is not always available, making the embedding results deviate from the actual meaning. At the same time, our mental world contains many concepts about worldly facts. For human cognition, compared to knowledge that we learned, common-sense concepts are more basic and general, and they play important roles in human knowledge accumulation. In this paper, based on common-sense concepts information of entities from a concept graph, we propose a Knowledge Graph Embedding with Concepts (KEC) model that embeds entities and concepts of entities jointly into a semantic space. The fact triplets from a knowledge graph are adjusted by the common-sense concept information of entities from a concept graph. Our model not only focuses on the relevance between entities but also focuses on their concepts. Thus, this model offers precise semantic embedding. We evaluate our method on the tasks of knowledge graph completion and entity classification. Experimental results show that our model outperforms other baselines on the two tasks.