Empowering Heterogeneous Graph Foundation Models via Decoupled Relation Alignment

2026-05-01

AI Summary

This work addresses the challenges of “type collapse” and “relation confusion” in graph foundation models operating on multi-domain heterogeneous graphs, which arise from cross-type feature shifts and intra-domain relational gaps. To mitigate these issues, we propose Decoupled Relation Subspace Alignment (DRSA), a novel relation-driven decoupled alignment framework. DRSA employs a dual relation subspace projection mechanism to disentangle feature semantics from relational structure and explicitly models cross-type interactions within a shared low-rank relation subspace. Additionally, it introduces decoupled feature-structure representations to accommodate intra-domain variations, thereby avoiding semantic distortion and topological disruption caused by global alignment. As a plug-and-play preprocessing module, DRSA consistently and significantly enhances the cross-domain and few-shot knowledge transfer performance of mainstream graph foundation models across multiple real-world benchmarks.

Abstract

While Graph Foundation Models (GFMs) have achieved remarkable success in homogeneous graphs, extending them to multi-domain heterogeneous graphs (MDHGs) remains a formidable challenge due to cross-type feature shifts and intra-domain relation gaps. Existing global feature alignment methods (PCA or SVD) enforce a shared feature space blindly, which distorts type-specific semantics and disrupts original topologies, inevitably leading to "Type Collapse" and "Relation Confusion". To address these fundamental limitations, we propose Decoupled relation Subspace Alignment (DRSA), a novel, plug-and-play relation-driven alignment framework. DRSA fundamentally shifts the paradigm by decoupling feature semantics from relation structures. Specifically, it introduces a dual-relation subspace projection mechanism to coordinate cross-type interactions within a shared low-rank relation subspace explicitly. Furthermore, a feature-structure decoupled representation is designed to decompose aligned features into a semantic projection component and a structural residual term, adaptively absorbing intra-domain variations. Optimized via a stable alternating minimization strategy based on Block Coordinate Descent, DRSA constructs a well-calibrated, structure-aware latent space. Extensive experiments on multiple real-world benchmark datasets demonstrate that DRSA can be seamlessly integrated as a universal preprocessing module, significantly and consistently enhancing the cross-domain and few-shot knowledge transfer capabilities of state-of-the-art GFMs. The code is available at: https://github.com/zhengziyu77/DSRA.

Problem

• Heterogeneous Graphs
• Graph Foundation Models
• Feature Alignment
• Relation Confusion
• Type Collapse

Innovation

• Decoupled relation Subspace Alignment
• Heterogeneous Graph Foundation Models
• Feature-Structure Decoupling
• Cross-Domain Alignment
• Relation Subspace Projection