Hyperbolic Geometry is Not Necessary: Lightweight Euclidean-Based Models for Low-Dimensional Knowledge Graph Embeddings

Recent knowledge graph embedding (KGE) models based on hyperbolic geometry\nhave shown great potential in a low-dimensional embedding space. However, the\nnecessity of hyperbolic space in KGE is still questionable, because the\ncalculation based on hyperbolic geometry is much more complicated than\nEuclidean operations. In this paper, based on the state-of-the-art\nhyperbolic-based model RotH, we develop two lightweight Euclidean-based models,\ncalled RotL and Rot2L. The RotL model simplifies the hyperbolic operations\nwhile keeping the flexible normalization effect. Utilizing a novel two-layer\nstacked transformation and based on RotL, the Rot2L model obtains an improved\nrepresentation capability, yet costs fewer parameters and calculations than\nRotH. The experiments on link prediction show that Rot2L achieves the\nstate-of-the-art performance on two widely-used datasets in low-dimensional\nknowledge graph embeddings. Furthermore, RotL achieves similar performance as\nRotH but only requires half of the training time.\n