Adaptive Distance Message Passing From the Multi-Relational Edge View

Zhongtian Sun, Alexandra I. Cristea, Pietro Lio, Jialin Yu

Keywords: Message passing, graph representation learning, multi-relations

TL;DR: Message of edges can be passed to nodes at different distances

Abstract: Message-passing graph neural networks (MP-GNNs) excel in deep learning on graphs. Despite their success in various studies, they are limited by passing information to the fixed length $k$ distance neighbouring nodes, where $k$ is the number of layers. In reality, different types of edges (alternatively relations) may influence nodes at varying distance and should not be uniformly treated. This paper proposes an adaptive distance message-passing method that considers the unique roles of edge types, addressing this issue. Experiments on real-world datasets validate the effectiveness of our approach.