Knowledge Graph Embedding Model with Attribute Information Fusion in Hybrid Space

The effectiveness of knowledge graph embeddings largely depends on the ability to model entity information and relational structures. However, most existing methods typically embed knowledge graphs into a single geometric space, resulting in embedding models that cannot simultaneously model chain-like, hierarchical, and cyclic structures. This paper proposes an embedding model, HMA, that integrates attribute information in mixed spaces. By leveraging interaction information from different geometric spaces and considering the differential features of these spaces, HMA captures the complex relational structures in knowledge graphs. Additionally, the fusion of triple relationship structures and entity attribute information obtained in mixed geometric spaces allows for the full utilization of entity information, thereby improving the quality of knowledge graph embeddings. We conducted extensive experiments on benchmark datasets such as YAGO3-10 and FB15k-237. The experimental results demonstrate that HMA outperforms many existing knowledge graph embedding models in terms of Hits@k and Mean Reciprocal Ranking metrics.