Go to ICML 2025 Conference homepageOpen Your Eyes: Vision Enhances Message Passing Neural Networks in Link Prediction
TL;DR: We first integrate vision structural awareness into MPNNs for link prediction, exploring new research questions and offering best practices.
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.
Lay Summary: A new approach called Graph Vision Network (GVN) is changing how we predict connections in networks, like social networks or web links. Typically, these predictions rely on analyzing the structure of the network, but GVN introduces a fresh perspective by using visual insights to improve accuracy. This means it looks at the network in a way similar to how we interpret visual information, making predictions more reliable. Tests across different types of networks show that GVN consistently outperforms older methods and sets new standards for accuracy. This development opens up exciting new possibilities for using visual understanding in network analysis, making it a promising area for future research and practical applications.
Primary Area: Deep Learning->Graph Neural Networks
Keywords: Multimodal Learning, Link Prediction, Graph Neural Networks
Submission Number: 1277
Loading