Incorporating Description Embeddings into Medical Knowledge Graphs Representation Learning

Abstract

Representation learning of medical knowledge graphs aims to embedding entities and relations in low-dimensional vector spaces, which is beneficial to the application of medical knowledge graphs in intelligent medical systems such as intelligent guidance, disease risk prediction and question answering system of medical field. Recently, some translation-based methods including TransE, TransH and TransR built entity and relation embeddings by regarding a relation as translation from head entity to tail entity. These methods solely use the information of triplets and don’t take text information into consideration. In this paper, we process a novel representation learning method by incorporating the embeddings of entity descriptions with classical translation-based methods. The embeddings of entity descriptions are built by Doc2Vec. It is easily applied for a large-scale domain-specific knowledge graphs because of its simplicity. Besides, we compare our method with classical translation-based methods to demonstrate the effectiveness of our method in medical knowledge graphs representation learning.