Learning Gaussian Hierarchy Embedding for Relation Prediction in Knowledge Graph

Abstract

Knowledge graph embedding methods, which transform high-dimensional and complex graph contents into low-dimensional vectors (or matrices, tensors) representations. They are useful for a wide range of graph analysis tasks including link prediction, node classification, recommendation and visualization. Most existing approaches represent graph nodes as vectors in a low-dimensional embedding space, ignoring the semantic hierarchy and the uncertainty present in the real-world graphs. To address this challenge, in this paper, we propose GHAKE, Gaussian Hierarchy Knowledge Graph Embedding, a novel graph embedding method that preserves both semantic hierarchy and node uncertain effectively and efficiently in an end-to-end manner. GHAKE effectively models uncertainty through Gaussian embeddings, and models semantic hierarchy via mapping entities into the polar coordinate system. In our comprehensive experiments, we evaluate GHAKE on real-world graphs, and the results demonstrate that can effectively model the semantic hierarchies and the uncertain of entities in knowledge graphs, and significantly outperforms existing state-of-the-art methods on benchmark datasets for the link prediction task.