Dual-View Embedding for Hyper-relational Knowledge Graphs with Hierarchical Structure

This study proposes a dual-view hyper-relational knowledge graph embedding model aimed at addressing the challenges of embedding complex relationships in knowledge graphs. Traditional methods primarily handle simple triplet relations and struggle with the complexity of hyper-relations. By integrating instance view and ontology view, our model, DVHE, captures hierarchical structural information between entities and is applied to link prediction tasks. Experimental results show that DVHE significantly outperforms existing single-view and dual-view models across multiple benchmark datasets, particularly in handling complex hyper-relations and hierarchical information. Ablation studies further validate the effectiveness of the model’s components, providing new insights for the development of knowledge graph embeddings.