Block-Diagonal Orthogonal Relation and Matrix Entity for Knowledge Graph Embedding

TL;DR

This paper introduces OrthogonalE, a novel knowledge graph embedding model using block-diagonal orthogonal matrices and Riemannian optimization, which improves flexibility and outperforms existing models with fewer parameters.

Contribution

OrthogonalE is the first KGE model to utilize block-diagonal orthogonal matrices with Riemannian optimization, enhancing model flexibility and generality.

Findings

OrthogonalE significantly outperforms state-of-the-art KGE models.

OrthogonalE reduces the number of relation parameters.

OrthogonalE demonstrates improved generalization in experiments.

Abstract

The primary aim of Knowledge Graph embeddings (KGE) is to learn low-dimensional representations of entities and relations for predicting missing facts. While rotation-based methods like RotatE and QuatE perform well in KGE, they face two challenges: limited model flexibility requiring proportional increases in relation size with entity dimension, and difficulties in generalizing the model for higher-dimensional rotations. To address these issues, we introduce OrthogonalE, a novel KGE model employing matrices for entities and block-diagonal orthogonal matrices with Riemannian optimization for relations. This approach enhances the generality and flexibility of KGE models. The experimental results indicate that our new KGE model, OrthogonalE, is both general and flexible, significantly outperforming state-of-the-art KGE models while substantially reducing the number of relation parameters.