Research map / Graph learning / Graph self-supervised learning
Contrastive Multi-View Representation Learning on Graphs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Contrast first-order adjacency and PPR diffusion views with node-graph pairs | Proposed graph representation learning method · Node and graph classification benchmarks | accuracy | Helped |
| Increase the number of views from two to three (ablation) | Proposed graph representation learning method · Cora, Citeseer, Pubmed | accuracy: Cora: 86.8 ± 0.5 -> 85.3 ± 0.5; Citeseer: 73.3 ± 0.5 -> 71.2 ± 0.7; Pubmed: 80.1 ± 0.7 -> 79.9 ± 0.6 | Hurt |
| Use node-graph contrastive pairs rather than global-global pairs (ablation) | Proposed graph representation learning method · Graph classification benchmarks | accuracy: MUTAG: 85.4 ± 2.8 -> 89.7 ± 1.1; Reddit-Binary: 80.8 ± 1.8 -> 84.5 ± 0.6 | Helped |
| Contrast multi-scale encodings (ablation) | Proposed graph representation learning method · Node and graph classification benchmarks | accuracy: Cora: 86.8 ± 0.5 -> 83.2 ± 0.9; MUTAG: 89.7 ± 1.1 -> 88.0 ± 0.8 | Hurt |
| Use Jensen-Shannon rather than other MI estimators (ablation) | Proposed graph representation learning method · Graph classification benchmarks | accuracy: MUTAG: NCE 82.2 ± 3.2; JSD 89.7 ± 1.1; NT-Xent 75.4 ± 7.8; DV 83.4 ± 1.9 | Helped |
| Use NCE rather than other MI estimators (ablation) | Proposed graph representation learning method · Node classification benchmarks | accuracy: Cora: NCE 85.8 ± 0.7, NT-Xent 86.8 ± 0.5; Citeseer: NCE 73.3 ± 0.5; Pubmed: NCE 80.1 ± 0.7 | Mixed |
| Replace PPR diffusion with heat diffusion alongside adjacency (ablation) | Proposed graph representation learning method · Node and graph classification benchmarks | accuracy: Cora: 86.8 ± 0.5 -> 86.4 ± 0.5; MUTAG: 89.7 ± 1.1 -> 85.0 ± 1.9 | Hurt |
| Replace sum pooling with mean pooling (ablation) | Proposed graph representation learning method · Graph classification benchmarks | accuracy: MUTAG: 89.7 ± 1.1 -> 88.8 ± 0.7; Reddit-Binary: 84.5 ± 0.6 -> 79.4 ± 0.5 | Hurt |
| Replace sum pooling with DiffPool (ablation) | Proposed graph representation learning method · Node and graph classification benchmarks | accuracy: Cora DiffPool-1: 84.6 ± 0.8; Cora DiffPool-2: 83.2 ± 0.9; MUTAG DiffPool-2: 88.0 ± 0.8 | Hurt |
| Use dedicated encoders for each view rather than sharing one encoder (ablation) | Proposed graph representation learning method · Node and graph classification benchmarks | accuracy: Cora: 86.2 ± 0.6 -> 86.8 ± 0.5; MUTAG: 82.8 ± 1.9 -> 89.7 ± 1.1 | Helped |
| Apply normalization or regularization during pre-training (ablation) | Proposed graph representation learning method · Downstream tasks | accuracy | Hurt |
| Increase batch size (ablation) | Proposed graph representation learning method · Graph classification benchmarks | accuracy | Helped |
Closest papers by meaning
- Adaptive Multi-view Graph Contrastive Learning via Fractional Continuous Dynamics (NeurIPS 2026 · similarity 0.90)
- Decoupled Self-supervised Learning for Graphs (NeurIPS 2022 · similarity 0.90)
- AutoGCL: Automated Graph Contrastive Learning via Learnable View Generators (AAAI 2022 · similarity 0.89)
- S3GCL: Spectral, Swift, Spatial Graph Contrastive Learning (ICML 2024 · similarity 0.89)
- Multi-View Graph Clustering via Node-Guided Contrastive Encoding (ICML 2025 · similarity 0.89)
- Equivalence is All: A Unified View for Self-supervised Graph Learning (ICML 2025 · similarity 0.88)
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.