Learning and Reasoning on Knowledge and Heterogeneous Graphs in the era of Graph Foundation and Large Language Models

Knowledge graphs and heterogeneous graphs represent systems with multiple entity and relation types, but useful models must balance expressiveness, scalable learning, and faithful reasoning. This tutorial establishes shared notation for typed heterogeneous graphs, temporal knowledge graphs, and event-based temporal heterogeneous graphs.

It formalises major task families—including knowledge graph completion, query answering, node and graph prediction, and their temporal variants—with particular attention to evaluation protocols and leakage. The survey then organises recent work on graph foundation models and on combining large language models with graph-structured knowledge.

A final strand covers learning over temporal heterogeneous graphs and temporal knowledge graphs, where models must forecast evolving facts without violating time. Together, these perspectives expose common design choices and open problems across transfer, grounding, provenance, and temporal reasoning.

Paper: https://doi.org/10.14428/esann/2026.es2026-4