Category-Embodied Knowledge Embedding

Abstract

Knowledge graph (KG) embedding, which transforms both the entities and relations into continuous low-dimensional continuous vector space, has attracted considerable research. A large amount of models have been proposed for knowledge graph embedding. However, most previous approaches only regard the knowledge graph as a set of triples, ignoring the categories of the entities. In this paper, we take advantages of category information by modelling the category-specific embedding. Specially, we see the interaction between the category embedding and KG embedding as a closed loop, in which the category embedding and KG embedding are promoted mutually. Triples along with their categories are represented in a unified framework, in which way the embedding of triples are category-aware. We evaluate our model on multiple real-world KGs, and it show impressive improvements on link prediction and triple classification compared with other baselines.