Neural Message Passing on Structural Interaction Graphs for Fully-Inductive Graph Neural Networks

2026-08-09

AI Summary

This work addresses the challenge of generalizing graph neural networks to unseen graphs under heterogeneity in feature dimensions, semantics, and structure. To this end, the authors propose SIGIL, a novel framework that reformulates the input graph into a structural interaction graph whose nodes correspond to feature dimensions, enabling a relation-based message-passing network to embed node features of arbitrary dimensionality into a unified representation space. The approach inherently satisfies equivariance with respect to permutations of nodes, features, and labels, thereby unifying the design paradigms of knowledge graph and general-purpose graph foundation models while rigorously generalizing existing knowledge graph reasoning architectures. Experimental results demonstrate that a single pre-trained SIGIL model achieves strong performance on fully inductive link prediction tasks, confirming its cross-graph transferability and architectural unification.

Abstract

A central obstacle in building graph foundation models is the input heterogeneity in terms of feature space dimensionality, semantics, and structure. Such heterogeneity limits the capability of graph neural networks to generalize to new graphs with unseen feature spaces. We address the transferability challenge with SIGIL, a framework that maps any attributed graph to a unified representation space of fixed dimension. Given a graph, SIGIL lifts it to a structural interaction graph, where nodes are the input feature dimensions and weighted, typed edges encode feature alignment across multiple orders of the graph's connectivity. A relational message-passing network embeds each feature dimension into a shared space, transforming the original node features, of arbitrary dimensionality, into representations transferable to any downstream graph. By construction, SIGIL is equivariant to permutations of nodes, feature dimensions, and labels. Additionally, when the input features are one-hot indicators of discrete relations, SIGIL recovers and strictly generalizes existing foundation models for knowledge graph reasoning. A single SIGIL model, pretrained on one graph, delivers strong fully-inductive link prediction. Also, SIGIL can be used to implement existing knowledge graph foundation models. As such, SIGIL unifies several existing regimes in graph foundation model design under a single framework

Problem

• graph neural networks
• input heterogeneity
• feature space
• transferability
• graph foundation models

Innovation

• Structural Interaction Graph
• Fully-Inductive GNN
• Relational Message Passing
• Feature Space Unification
• Graph Foundation Model