The idea in plain language.
Different relationship types should influence how neighboring entities exchange information.
How it works
An ordinary graph model may treat every connection in the same way. R-GCNs use relation-aware transformations when aggregating information from neighboring entities. Multiple layers let a representation incorporate evidence from a wider neighborhood. The paper applies these representations to predicting entity categories and completing missing links in knowledge graphs.
What to keep in mind
Results depend on the graph, relation types, and evaluation protocol. Predicting a plausible missing link does not independently verify that a fact is true.
Source: Modeling Relational Data with Graph Convolutional Networks. The original manuscript contains the methods, experiments, figures, and references. An arXiv posting date may follow an earlier conference publication. Read the linked record for version history.