Learning hyperbolic attention-based embeddings for link prediction in knowledge graphs
作者：
Highlights：
• Euclidean space cannot accurately preserve the hierarchies present in KGs.
• Hyperbolic space requires few dimensions to embed the hierarchical structures.
• An HGNN based encoder is used to enrich node embedding with neighborhood information.
• A recent hyperbolic model is used to predict new valid triples for KG completion.
• The results prove that richer hyperbolic embeddings lead to performance improvement.
摘要
•Euclidean space cannot accurately preserve the hierarchies present in KGs.•Hyperbolic space requires few dimensions to embed the hierarchical structures.•An HGNN based encoder is used to enrich node embedding with neighborhood information.•A recent hyperbolic model is used to predict new valid triples for KG completion.•The results prove that richer hyperbolic embeddings lead to performance improvement.
论文关键词：Knowledge graph,Link prediction,Graph neural network,Hyperbolic graph neural network,Poincaré ball model
论文评审过程：Received 24 September 2020, Revised 1 August 2021, Accepted 3 August 2021, Available online 5 August 2021, Version of Record 12 August 2021.
论文官网地址：https://doi.org/10.1016/j.knosys.2021.107369
