Representation Learning for Knowledge Graph with Dynamic Step

Abstract

Representation learning aims to represent the entities and relations in a knowledge graph as dense, low-dimensional and real-valued vectors by machine learning. The translation-based model is a typical representation learning method and has shown good predictive performance in large-scale knowledge graph. However, when modeling complex relations such as 1-N, N-1 and N-N, these models are not very effective. To solve the limitation of traditional learning model in modeling complex relations, a representation learning method based on dynamic step is proposed. Defining a dynamic step according to the different types of relations can significantly improve the efficiency of learning. The algorithm is used to solve the problem of single optimization goal, and the experimental results show that the dynamic step method can mainly improve the performance in the link prediction task.