Knowledge Graph Embedding using Tri-Relation Vectors on Hyperplanes

Graph neural networks are a subfield of graph learning within artificial intelligence. Knowledge graphs enhance information interconnectivity and comprehensibility by representing knowledge in a structured and semantic manner. They provide richer semantic information compared to traditional datasets by explicitly modeling relations and entities. Knowledge graphs play a key role in powering search engines, answering questions, and providing recommendations. A major focus of research in this area is link prediction, which aims to uncover missing relationships between entities that have not yet been recorded. Knowledge graph embedding, a highly efficient and scalable approach for link prediction. This paper focuses on distance-based models for KGE, which are lightweight, easy to train, and geometrically interpretable. We propose a new model, TriRH, which uses tri-relation vectors on hyperplanes to enhance predictive accuracy and overall model performance. Experiments on benchmark datasets show that TriRH achieves better performance than existing models across multiple metrics.