Exploring the generalization of knowledge graph embedding

Abstract

Knowledge graph embedding aims to represent structured entities and relations as continuous and dense low-dimensional vectors. With more and more embedding models being proposed, it has been widely used in many tasks such as semantic search, knowledge graph completion and intelligent question and answer. Most knowledge graph embedding models focus on how to get information about different entities and relations. However, the generalization of knowledge graph embedding or the link prediction ability is not well-studied empirically and theoretically. The study of generalization ability is conducive to further improving the performance of the model. In this paper, we propose two measures to quantify the generalization ability of knowledge graph embedding and use them to analyze the performance of translation-based models. Extensive experimental results show that our measures can well evaluate the generalization ability of a knowledge graph embedding model.