Graphs, Entities, and Step Mixture

TL;DR

The paper introduces GESM, a novel graph neural network that mitigates oversmoothing and improves generalization by combining edge relationships, entity features, and step mixture via random walk, achieving state-of-the-art results.

Contribution

It proposes GESM, a new GNN model that integrates edge and node features with random walk step mixture to address oversmoothing and generalization issues.

Findings

GESM achieves state-of-the-art or comparable performance on eight benchmark datasets.

The model effectively alleviates oversmoothing through step mixture and attention mechanisms.

Global information consideration significantly improves GNN performance.

Abstract

Existing approaches for graph neural networks commonly suffer from the oversmoothing issue, regardless of how neighborhoods are aggregated. Most methods also focus on transductive scenarios for fixed graphs, leading to poor generalization for unseen graphs. To address these issues, we propose a new graph neural network that considers both edge-based neighborhood relationships and node-based entity features, i.e. Graph Entities with Step Mixture via random walk (GESM). GESM employs a mixture of various steps through random walk to alleviate the oversmoothing problem, attention to dynamically reflect interrelations depending on node information, and structure-based regularization to enhance embedding representation. With intensive experiments, we show that the proposed GESM achieves state-of-the-art or comparable performances on eight benchmark graph datasets comprising transductive and inductive learning tasks. Furthermore, we empirically demonstrate the significance of considering global information.