Knowledge Representation Learning with Hyperboloid Models

Knowledge graph describes the objective knowledge in the real world in a structured form. Knowledge representation learning aims to project the entities and relations in the knowledge graph into a low-dimensional, continuous vector space and learn vector representations of the semantic information of entities and relations. Moreover, link prediction and other tasks are carried out by vector calculation, and a variety of downstream tasks are achieved while retaining the structure of the knowledge graph itself. However, the existing knowledge representation learning models have certain defects when modeling complex relationships such as 1-N, N-1, and N-N, and the processing effect is not well. In order to better deal with complex relations, this paper proposes a knowledge representation learning model based on hyperboloid model, which emitters the knowledge graph into hyperbolic space for knowledge representation learning. Experiments are carried out on the public datasets FB15K, FB15K-237 and WN18RR. The experimental results show that the proposed model is superior to the baseline model in most evaluation indicators, which proves effectiveness of the proposed model.