FITTER Enables Vocabulary-Agnostic Inference on Temporal Knowledge Graphs
August 12, 2026
FITTER is an inductive structural model for temporal knowledge graph link prediction that supports cross-domain transfer. It utilizes relative temporal encodings and message-passing to generate embeddings for entirely unseen entities and relations without requiring training-time vocabulary.
HOW THIS AFFECTS YOU
●
researcherYou can apply this model to temporal graphs containing entities and relations not seen during the training phase.