QUARTET: Enhancing Graph Transformers via Quad-branch Cross-Attention

QUARTET improves relational deep learning by replacing local random samplers with Causal Random Walk sampling and using quad-branch cross-attention for global context. This architecture addresses connectivity issues and memory limitations found in models like RelGT when processing heterogeneous temporal graphs.