Research map / Graph learning / Graph algorithms and clustering
LoNe Sampler: Graph Node Embeddings by Coordinated Local Neighborhood Sampling
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 |
|---|---|---|---|
| Use L1 or L2 coordinated sampling instead of NodeSketch | node classification embeddings · Cora, Citeseer, PubMed, DeezerEurope, LastFM, GitWebML | accuracy: Cora: 0.833 vs 0.831; Citeseer: 0.706 -> 0.722; PubMed: 0.798 -> 0.818; DeezerEurope: 0.550 -> 0.565; LastFM: 0.809 -> 0 | Mixed |
| Use L0 sampling rather than L1/L2 sampling | LoNe Sampler node classification embeddings · Cora, Citeseer, PubMed, DeezerEurope, LastFM, GitWebML | accuracy: 0.824 vs 0.831/0.829; 0.702 vs 0.711/0.722; 0.781 vs 0.818/0.804; 0.554 vs 0.558/0.565; 0.725 vs 0.825/0.818; 0.831 vs 0 | Mixed |
| Use LoNe Sampler L1/L2 instead of DeepWalk or node2vec | node classification embeddings · Cora, Citeseer, PubMed, DeezerEurope | accuracy: Cora L1 0.831 vs node2vec 0.815; Citeseer L2 0.722 vs 0.572; PubMed L1 0.818 vs 0.807; DeezerEurope L2 0.565 vs 0.561 | Helped |
| Use LoNe Sampler L1/L2 instead of node2vec | node classification embeddings · LastFM and GitWebML | accuracy: LastFM: 0.866 -> 0.825; GitWebML: 0.863 -> 0.837 | Hurt |
| Increase embedding dimensionality | LoNe Sampler · node classification and link prediction | performance | Helped |
| Increase L1/L2 sketch size | LoNe Sampler · node classification | performance | Mixed |
| Sample larger neighborhoods with L0 | LoNe Sampler L0 · node classification | accuracy | Hurt |
| Use larger neighborhoods for link prediction rather than hop-1 sampling | LoNe Sampler · link prediction | precision@1000 and recall@1000 | Hurt |
| Use LoNe Sampler rather than NodeSketch | link prediction · PubMed and GitWebML | precision@1000 and recall@1000 | Helped |
| Use LoNe Sampler rather than NodeSketch | link prediction · DeezerEurope | precision@1000 and recall@1000: < 0.001 | No clear effect |
| Replace a precomputed kernel SVM with the explicit feature-map linear SVM | NodeSketch embeddings · GitWebML SVM training and inference | runtime: OOM -> 15.9 | Helped |
| Use more parallel threads | LoNe Sampler L1, 50-dimensional embeddings · Cora, Citeseer, PubMed, Deezer, LastFM, GitWebML | runtime: Cora: 25.7 -> 5.4; Citeseer: 30.4 -> 6.3; PubMed: 214.3 -> 42.2; Deezer: 344.4 -> 62.9; LastFM: 128.5 -> 26.5; GitWebML: | Helped |
Closest papers by meaning
- Node Representation Learning in Graph via Node-to-Neighbourhood Mutual Information Maximization (CVPR 2022 · similarity 0.90)
- Reconciling Competing Sampling Strategies of Network Embedding (NeurIPS 2023 · similarity 0.89)
- Learning Representations for Hierarchies with Minimal Support (NeurIPS 2024 · similarity 0.88)
- vGraph: A Generative Model for Joint Community Detection and Node Representation Learning (NeurIPS 2019 · similarity 0.88)
- Community Detection Guarantees using Embeddings Learned by Node2Vec (NeurIPS 2024 · similarity 0.88)
- Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning (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.