GraphShaper: Geometry-aware Alignment for Improving Transfer Learning in Text-Attributed Graphs

2025-10-13

AI Summary

Existing graph representation learning methods predominantly rely on a single Euclidean space, failing to simultaneously accommodate the intrinsic geometric disparities between tree-like structures (naturally modeled in hyperbolic geometry) and cycle-rich structures (better captured by spherical geometry). This mismatch induces geometric constraint conflicts at structural boundaries, degrading cross-structural transfer performance. To address this, we propose Multi-Geometry Expert Network (MG-Net), which jointly models hyperbolic and hyperspherical geometries in parallel, enhances semantic awareness via large language model integration, and introduces a geometry-adaptive weight fusion mechanism to achieve structure-sensitive representation alignment. Evaluated on zero-shot cross-structural transfer tasks, MG-Net achieves a 9.47% accuracy gain on citation networks and a 7.63% improvement on social networks, demonstrating substantially enhanced generalization capability over heterogeneous graph topologies.

Technology Category

• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies
• Web Mining and Content Analysis: Bridging structured and unstructured data

Abstract

Graph foundation models represent a transformative paradigm for learning transferable representations across diverse graph domains. Recent methods leverage large language models to unify graph and text modalities into a shared representation space using contrastive learning. However, systematic evaluations reveal significant performance degradation at structural boundaries where distinct topological patterns converge, with accuracy losses exceeding 20 percentage points. This issue arises from a key limitation: current methods assume all graph structures can be encoded within a single Euclidean space. In reality, tree structures require hyperbolic geometry to preserve hierarchical branching, while cyclic patterns depend on spherical geometry for closure properties. At structural boundaries, nodes experience conflicting geometric constraints that uniform encoding spaces cannot resolve. This raises a crucial challenge: extbf{Can alignment frameworks be designed to respect the intrinsic geometric diversity of graph structures?} We introduce extbf{GraphShaper}, a geometry-aware framework that enhances graph encoding through multi-geometric specialization. Our approach employs expert networks tailored to different geometric spaces, dynamically computing fusion weights to adaptively integrate geometric properties based on local structural characteristics. This adaptive fusion preserves structural integrity before alignment with text embeddings. Extensive experiments demonstrate that GraphShaper achieves 9.47% accuracy improvements on citation networks and 7.63% on social networks in zero-shot settings.

Problem

• Addresses performance degradation at structural boundaries in graph learning
• Resolves conflicting geometric constraints in heterogeneous graph structures
• Enables geometry-aware alignment between graph and text modalities

Innovation

• Employs multi-geometric expert networks for graph encoding
• Dynamically fuses geometric properties based on local structures
• Aligns geometry-preserved graph embeddings with text embeddings