KGPFN: Unlocking the Potential of Knowledge Graph Foundation Model via In-Context Learning

Researchers introduced KGPFN, a knowledge graph foundation model that combines relation-level pretraining with local and global contextual evidence.

The model uses in-context learning to improve graph reasoning, especially on unseen graphs, entities, and relations.

Across 57 benchmarks, KGPFN reported strong results and outperformed or matched prior approaches in many settings.

The work suggests foundation models for knowledge graphs can generalize better without depending only on fine-tuning.