Survey on Knowledge Graph Embedding Based on Hyperbolic Geometry
Mengqi Lin, Qing Min Liao, Yan Jia, Ye Wang · 2021
It is a prevalence to store human knowledge in knowledge graphs that connect entities with relations. Recently, temporal knowledge graph has received more attention, since real-world events are dynamic. The representation learning of temporal knowledge graph can better satisfy the needs of practical applications. Traditional Euclidean embedding methods are shown to incur large distortion when represent hierarchical or scale-free network. Luckily, hyperbolic geometry provides powerful tools for tackling this problem and is capable of efficiently representing knowledge graphs with these properties. In this survey, we summarize and compare the existing work of the Euclidean embeddings and the hyperbolic embeddings, and demonstrate temporal knowledge graph embedding based on hyperbolic geometry is a promising research direction.