DGKT: Dual Graph Kernel Transformer for graph classification

Lixiang Xu

doi:10.1016/j.patcog.2026.115145

DGKT: Dual Graph Kernel Transformer for graph classification

2026-10-16

0

PRE

AI

DOI：10.1016/j.patcog.2026.115145

Original

Original request for help

Save

Abstract

En 中文

• We derive differentiable Weisfeiler–Lehman and random-walk kernel encodings. • We propose DGKT with hybrid local modeling, dual kernels, and dual-branch learning. • We adaptively initialize dual-kernel branches using node-wise structural importance. • We compare favorably with GNNs, Graph Transformers, and kernels on seven datasets.

Keywords:

Graph classification

Graph transformer

Graph kernels

Positional encoding

Graph representation learning

DGKT: Dual Graph Kernel Transformer for graph classification

Abstract

Journal

Organization

Cited Papers