Abstract: Aimed at the current knowledge graph embedding methods that mostly focus on the entity and relationship information in the triples, thus ignoring the rich information related to the relationships other than the triples, a knowledge graph embedding method, CompGCN-RHS, fused with relational hierarchical structure information is proposed. The hierarchical structure information of the relationship was incorporated into the representation of the relationship, the entity and the relationship were combined for embedded learning, and the attention mechanism was introduced to learn the different contributions of different neighbor nodes to the central node when the neighbor node information was aggregated. On the dataset Sport, the MRR and Hits@1 of this method were increased by 2.2 percentage points and 2.3 percentage points respectively; on the Location, they are increased by 4.7 percentage points and 6 percentage points respectively. The experimental results verified the effectiveness of the method.