A Knowledge Reasoning Algorithm Based on Network Structure and Representation Learning

Representation learning of knowledge graphs is widely used in the field of artificial intelligence, but it is currently far from comprehensive. All previous translation-based models ignore the multi-relational network structure attributes. Inspired by the law of gravitation, a novel algorithm strTransE proposed by us to integrate the network structure attributes into representation learning. In strTransE, the relation is the gravity between entities. The magnitude of gravity is determined by the weight of the entity. In order to solve the gravitational quantification, we introduced the network structure attribute to represent the weight of the entity. Experimental results on datasets WN18 and FB15K show that, our algorithm achieves significant and continuous improvements in link prediction.