Research map / Graph learning / Graph self-supervised learning
Inductive Representation Learning on Large 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 |
|---|---|---|---|
| replace the convolutional aggregator with trainable LSTM or pooling aggregation (ablation) | GraphSAGE · Citation, Reddit, and PPI; supervised and unsupervised | micro-averaged F1: 7.4% on average | Helped |
| add concatenation of the node representation and aggregated neighborhood representation (skip connection) (ablation) | GraphSAGE mean aggregator · Citation, Reddit, and PPI | micro-averaged F1 | Helped |
| increase aggregation depth from K=1 to K=2 (ablation) | GraphSAGE · node-classification benchmarks | accuracy: 10-15% | Helped |
| increase aggregation depth beyond K=2 (ablation) | GraphSAGE · node-classification benchmarks | accuracy: 0-5% | Mixed |
| use max-pooling instead of mean-pooling in the pooling aggregator (ablation) | GraphSAGE-pool · development tests | performance | No clear effect |
| use pooling rather than mean aggregation (ablation) | GraphSAGE · Citation, Reddit, and PPI; supervised and unsupervised | micro-averaged F1 | Mixed |
| use LSTM rather than mean aggregation (ablation) | GraphSAGE · Citation, Reddit, and PPI; supervised and unsupervised | micro-averaged F1 | Helped |
| use pooling rather than LSTM aggregation (ablation) | GraphSAGE · Citation, Reddit, and PPI; supervised and unsupervised | micro-averaged F1: T=10.0, p=0.46 | No clear effect |
| use GraphSAGE-pool instead of DeepWalk + features | GraphSAGE-pool · Citation and Reddit; supervised and unsupervised | micro-averaged F1: 13.8%; 29.1%; 19.7%; 37.2% | Helped |
| use GraphSAGE instead of raw node features alone | GraphSAGE · Citation, Reddit, and PPI | micro-averaged F1: 39%; 46%; 55%; 63%; 19%; 45% | Helped |
| use GraphSAGE rather than DeepWalk for embedding unseen test nodes | GraphSAGE · Reddit test set | test-time runtime: 100-500× | Helped |
| sample fixed-size neighborhoods rather than use full neighborhoods (ablation) | GraphSAGE · node-classification benchmarks | accuracy | No clear effect |
Closest papers by meaning
- Watch Your Step: Learning Node Embeddings via Graph Attention (NeurIPS 2018 · similarity 0.90)
- Deep Gaussian Embedding of Graphs: Unsupervised Inductive Learning via Ranking (ICLR 2018 · similarity 0.89)
- iN2V: Bringing Transductive Node Embeddings to Inductive Graphs (ICML 2025 · similarity 0.89)
- Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning (ICLR 2025 · similarity 0.88)
- Learning Representations for Hierarchies with Minimal Support (NeurIPS 2024 · similarity 0.88)
- Fully-inductive Node Classification on Arbitrary Graphs (ICLR 2025 · similarity 0.87)
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.