.png){kind=icon}
.png){kind=icon}
Return
HYPER2: Hyperbolic embedding for hyper-relational link prediction
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2022.04.026.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Knowledge graphs (KGs) Embedding has been broadly studied in recent years. However, less light is shed on the ubiquitous hyper-relational KGs. Most existing hyper-relational KG embedding methods decompose n-ary facts into smaller tuples, undermining the structure of n-ary facts. Moreover, these models always suffer from low expressiveness and high complexity. In this work, to tackle the indecomposability issue, we represent n-ary fact as a hyperedge, keeping the integrity of fact and maintaining the vital role that primary triple plays. To address the expressiveness and complexity issue, we propose HYPER2 where we generalize hyperbolic Poincare embedding from binary to arbitrary arity data, and we design an information aggregation module to capture the interaction between entities within and beyond triple. Extensive experiments demonstrate HYPER2 is superior to its translational and deep analogues, improving MRR and other metrics by a large margin with relatively few dimensions. Moreover, we study the side effect of literals, and we theoretically and experimentally compare the computational complexity of HYPER2 against several best-performing baselines. HYPER2 is much quicker than its counterparts. (C) 2022 Published by Elsevier B.V.
Keywords:
Link prediction
Knowledge graph embedding
Hyperbolic space
Hyper-relational knowledge graph
Hyper-relational link prediction