Research map / Learning theory and algorithms / Clustering
Community models for networks observed through edge nominations
Tianxi Li, Elizaveta Levina, Ji Zhu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 right singular vectors (Right SC) instead of left, spherical-left or symmetrized spectral clustering (ablation) | community detection under NSBM · Synthetic unweighted NSBM, n=1200, K=3, beta=0.2, t from 0.2 to 2 | cluster accuracy | Improved |
| increase nomination heterogeneity t with symmetric clustering methods | Symmetric SC / SSC · Synthetic unweighted NSBM, beta=0.2 | cluster accuracy | Hurt |
| use the NSBM estimator instead of SBM, DCSBM, SCBM or Bayesian model for probability matrix | estimation of probability matrix · Synthetic unweighted NSBM, beta=0.2 | relative estimation error of P-tilde | Improved |
| use Bayesian missing-link model before spectral clustering | Bayesian-Left/Symmetric/Right · Synthetic unweighted NSBM | cluster accuracy | No clear effect |
| increase beta (weaker community separation) to 0.7 or more | Right SC · Synthetic unweighted NSBM, t=1.5 | cluster accuracy: all methods uninformative for beta >= 0.7 | Hurt |
| use NSBM instead of simpler SBM and SCBM at large beta | probability matrix estimation · Synthetic unweighted NSBM, t=1.5 | relative estimation error of P-tilde: better for beta <= 0.6, worse for larger beta | Mixed |
| use Poisson edge weights instead of Bernoulli edges | Right SC · Synthetic weighted NSBM, beta=0.2 | cluster accuracy: accurate for t from 0 to 2; stable until beta > 0.8 | Improved |
| cluster with right singular vectors (NSBM) instead of symmetric spectral clustering | business school hiring network · 87 US business schools, K=4 | average US News ranking by community: 7.7, 29.8, 53.1, 63.7 vs 19.2, 55.1, 52.7, 63.8 | Improved |
| truncate hiring counts to weights 1 and 2 instead of binary 0/1 (ablation) | hiring network construction · 87 US business schools | stability | Improved |
Closest papers by meaning
- Analysis of spectral clustering algorithms for community detection: the general bipartite setting (JMLR 2019 · similarity 0.91)
- Achieving Optimal Misclassification Proportion in Stochastic Block Models (JMLR 2017 · similarity 0.90)
- Recovering Unbalanced Communities in the Stochastic Block Model with Application to Clustering with a Faulty Oracle (NeurIPS 2023 · similarity 0.90)
- Individual-centered Partial Information in Social Networks (JMLR 2024 · similarity 0.89)
- Supervised Community Detection with Line Graph Neural Networks (ICLR 2019 · similarity 0.89)
- CataBEEM: Integrating Latent Interaction Categories in Node-wise Community Detection Models for Network Data (ICML 2023 · similarity 0.89)
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.
