Autoregressive Models for Knowledge Graph Generation

KG semantic dependency generation: Knowledge graph generation requires learning complex semantic dependencies between triples while maintaining domain validity constraints.

Subgraph interdependency modeling: Generative KG models must capture interdependencies across entire subgraphs to produce semantically coherent structures, unlike independent triple scoring in link prediction.

KG semantic dependency generation: Knowledge graph generation requires learning complex semantic dependencies between triples while maintaining domain validity constraints.

Subgraph interdependency modeling: Generative KG models must capture interdependencies across entire subgraphs to produce semantically coherent structures, unlike independent triple scoring in link prediction.

ARK autoregressive KG generator: ARK is a family of autoregressive models that generates knowledge graphs by treating graphs as sequences of head-relation-tail triples.

Implicit semantic constraint learning: ARK learns semantic constraints such as type consistency, temporal validity, and relational patterns directly from data without explicit rule supervision.

SAIL variational controlled generation: SAIL is a variational extension of ARK for controlled generation through learned latent representations, supporting unconditional sampling and conditional completion.

ARK autoregressive KG generator: ARK is a family of autoregressive models that generates knowledge graphs by treating graphs as sequences of head-relation-tail triples.

Implicit semantic constraint learning: ARK learns semantic constraints such as type consistency, temporal validity, and relational patterns directly from data without explicit rule supervision.

High semantic validity:

Capacity more important than depth:

Recurrent efficiency with comparable validity: