Open the Eyes of MPNN: Vision Enhances MPNN in Link Prediction

2025-05-13

AI Summary

This work addresses the limitation of message-passing neural networks (MPNNs) in graph link prediction—namely, their neglect of visual structural information. We introduce, for the first time, a vision-aware mechanism into the MPNN framework, proposing the Graph Vision Network (GVN) and its lightweight variant, E-GVN. Methodologically, GVN integrates visual feature encoding, structure-aware attention, and efficient graph message passing, enabling end-to-end training and plug-and-play compatibility with mainstream MPNN architectures. Our core contribution is the establishment of a novel “graph vision enhancement” paradigm, endowing MPNNs with explicit visual-structural perception capability. Extensive experiments on seven standard link prediction benchmarks—including large-scale graphs—demonstrate that GVN consistently outperforms existing state-of-the-art methods, achieving new best-in-class performance across all datasets.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Computer Vision: Multi-modal Vision
• 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: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Web data visualization

Abstract

Message-passing graph neural networks (MPNNs) and structural features (SFs) are cornerstones for the link prediction task. However, as a common and intuitive mode of understanding, the potential of visual perception has been overlooked in the MPNN community. For the first time, we equip MPNNs with vision structural awareness by proposing an effective framework called Graph Vision Network (GVN), along with a more efficient variant (E-GVN). Extensive empirical results demonstrate that with the proposed frameworks, GVN consistently benefits from the vision enhancement across seven link prediction datasets, including challenging large-scale graphs. Such improvements are compatible with existing state-of-the-art (SOTA) methods and GVNs achieve new SOTA results, thereby underscoring a promising novel direction for link prediction.

Problem

• Enhance MPNNs with visual perception for link prediction
• Propose Graph Vision Network (GVN) for vision-aware MPNNs
• Achieve SOTA results by integrating vision with structural features

Innovation

• Enhance MPNNs with vision structural awareness
• Propose Graph Vision Network (GVN) framework
• Achieve SOTA results in link prediction