Knowledge Graph Embedding Model Based on Three-Dimensional Spatial Hierarchy Preservation

Knowledge graph embedding is a key technique that facilitates efficient execution of downstream tasks like link prediction and entity alignment. As knowledge graphs are increasingly applied in diverse domains, this technique holds significant value for enhancing their utility and functionality. However, existing knowledge graph embedding models often overlook the hierarchical relationships between entities. To address this issue, this paper proposes a novel model based on Three-Dimensional Spatial Hierarchy Preservation called 3DH- KGE. 3DH-KGE constructs a three-dimensional spatial representation to capture the multi-level relationships between entities. Relations are modeled as rotations, scalings, and translations in the three-dimensional space, while entities are represented as vectors in the same space. By incorporating scaling into the relation modeling, 3DH-KGE can represent entities at different hierarchical levels effectively. Experimental results on link prediction tasks demonstrate that the 3DH-KGE model achieves notable improvements in Mean Reciprocal Rank (MRR) and Hits@N on the WN18RR and FB15k-237 datasets, indicating that 3DH-KGE enhances the accuracy of link prediction tasks.