arxivcs.LG2026-07-10
Temporal Knowledge Graph Forecasting under Distribution Shifts: A Synthetic Evaluation
Konrad Özdemir, Julia Gastinger, Lukas Kirchdorfer, Heiner Stuckenschmidt
Temporal knowledge graphs (TKGs) represent evolving relational systems, whose underlying data-generating processes often change over time. Yet, TKG forecasting models are commonly evaluated only on empirical benchmark datasets that provide limited insight into the models' robustn…