Research map / Graph learning / Graph self-supervised learning
On the Benefits of Attribute-Driven Graph Domain Adaptation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes GAA, a cross-channel graph domain adaptation method that aligns both node-attribute (kNN feature graph) and topology views, motivated by a PAC-Bayes bound showing attribute divergence matters for GDA.
Problem
Unsupervised graph domain adaptation for cross-network node classification, where prior methods align only topological structure and ignore node attribute shift, which the paper argues is larger.
Approach
Builds a kNN feature graph from node attributes alongside the original topology graph, uses two GCN channels plus attention and a cross-view cosine similarity matrix to refine each view, and aligns both views between source and target graphs, with a PAC-Bayes argument motivating attribute alignment.
Key results
- GAA achieves the best accuracy on most Airport, Citation, Blog and Twitch tasks, e.g., 0.704 on U→B airport and 0.824 on A→C citation versus JHGDA's 0.695 and 0.814.
- Reported maximum average ACC improvement over the best competing method of 1.80%.
- On MAG, GAA slightly beats PA on most tasks (e.g., US→CN 0.410 vs 0.400), while UDAGCN, EGI and JHGDA run out of memory.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use GAA instead of the strongest baseline | Cross-network node classification · Airport U→B | accuracy: 0.695 -> 0.704 | Helped |
| Use GAA instead of the strongest baseline | Cross-network node classification · Airport E→B | accuracy: 0.740 -> 0.691 | Hurt |
| Use GAA instead of the strongest baseline | Cross-network node classification · Citation A→D | accuracy: 0.762 -> 0.789 | Helped |
| Use GAA instead of the strongest baseline | Cross-network node classification · Blog B1→B2 | accuracy: 0.662 -> 0.681 | Helped |
| Use GAA instead of the strongest baseline | Cross-network node classification · Twitch EN→DE | accuracy: 0.760 -> 0.751 | Hurt |
| Use GAA instead of the strongest baseline | Cross-network node classification · MAG US→CN | accuracy: 0.400 -> 0.410 | Helped |
| Add cross-view similarity matrix refinement (ablation) | GAA · Citation and Airport datasets | accuracy | Helped |
| Add attribute-driven alignment loss (ablation) | GAA · Citation and Airport datasets | accuracy | Helped |
| Add the feature graph and attribute-driven alignment (ablation) | GAA · Citation and Airport datasets | accuracy | Helped |
| Use a feature kNN graph instead of the original graph | GCN · Airport and Citation datasets | accuracy: +5.30%; +0.60% | Helped |
| Use GAA instead of the strongest baseline across reported tasks | Cross-network node classification · Four dataset groups | accuracy: 1.80% | Mixed |
Surprising findings
- Attribute-based kNN-GCN outperforms plain GCN by over 5% on average on airport data, supporting the claim that attribute views help.
Limitations
- Theoretical bound relies on a Gaussian class-conditional assumption and loose big-O terms.
- Ablation and tightness experiments are limited; some reported numbers are inconsistent in the paper (e.g., dataset naming and DBLP/Citation labels).
- Gains over strongest baselines (PA, JHGDA) are modest, often within a few points and single-seed-like averages.
Experimental setup
Datasets: Airport (USA, Brazil, Europe) Citation (ACMv9, Citationv1, DBLPv7) Blog (Blog1, Blog2) Twitch (Germany, England) MAG (US, CN, DE, JP, RU, FR)
Benchmarks: Airport Citation Blog Twitch MAG
Metrics: accuracy
Baselines: GCN kNN-GCN DANN DANE UDAGCN ASN EGI GRADE-N JHGDA SpecReg GIFI PA
Builds on: GCN DANN UDAGCN JHGDA
Closest papers by meaning
- Revisiting, Benchmarking and Understanding Unsupervised Graph Domain Adaptation (NeurIPS 2024 · similarity 0.91)
- Homophily Enhanced Graph Domain Adaptation (ICML 2025 · similarity 0.91)
- Rethinking Propagation for Unsupervised Graph Domain Adaptation (AAAI 2024 · similarity 0.90)
- Pairwise Alignment Improves Graph Domain Adaptation (ICML 2024 · similarity 0.90)
- Open-Set Graph Domain Adaptation via Separate Domain Alignment (AAAI 2024 · similarity 0.90)
- TFGDA: Exploring Topology and Feature Alignment in Semi-supervised Graph Domain Adaptation through Robust Clustering (NeurIPS 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.