Graph Domain Adaptation via Homophily-Agnostic Reconstructing Structure

TL;DR

This paper introduces a homophily-agnostic graph domain adaptation method that reconstructs and aligns different variants of source and target graphs, improving transfer learning especially on heterophilic graphs.

Contribution

It proposes a novel divide-and-conquer approach that reconstructs homophilic and heterophilic graph variants and aligns them separately, addressing the limitations of existing methods.

Findings

Outperforms existing GDA methods on five benchmark datasets.

Shows significant advantages on heterophilic graphs.

Effectively handles graphs with varying degrees of homophily.

Abstract

Graph Domain Adaptation (GDA) transfers knowledge from labeled source graphs to unlabeled target graphs, addressing the challenge of label scarcity. However, existing GDA methods typically assume that both source and target graphs exhibit homophily, leading existing methods to perform poorly when heterophily is present. Furthermore, the lack of labels in the target graph makes it impossible to assess its homophily level beforehand. To address this challenge, we propose a novel homophily-agnostic approach that effectively transfers knowledge between graphs with varying degrees of homophily. Specifically, we adopt a divide-and-conquer strategy that first separately reconstructs highly homophilic and heterophilic variants of both the source and target graphs, and then performs knowledge alignment separately between corresponding graph variants. Extensive experiments conducted on five benchmark datasets demonstrate the superior performance of our approach, particularly highlighting its substantial advantages on heterophilic graphs.