Research Progress of Knowledge Graph Embedding

Abstract

As an effective and efficient approach for describing real-world entities with relations, Knowledge graph (KG) technology has received a lot of attention in recent years. KG embedding technology becomes one hot spot in KG research in terms of its efficiency in KG completion, relationship extraction, entity classification and resolution, etc. KG embedding is to embed KG entities and relations as vectors into the continuous spaces, aims at simplifying the operation, also retaining the inherent structure of the KG. In this paper, we focus on the progress of KG embedding in recent these years and summarize the latest embedding methods systematically into a review. We classify these embedding methods into two broad categories based on the type of information in the KG used in this review. We started at introducing models that only use the fact information in the KG followed by briefly review of previous models, focusing on the new models and the improved models based on the previous models. We describe the core ideas, implementation methods, the advantages and disadvantages of these technologies. Then, models incorporate additional information, including entity types, relational paths, textual descriptions, and the use of logical rules will be also included.