NetPTR enables private community detection on sparse networks
New algorithm protects edge privacy while maintaining accuracy in network clustering...
NetPTR addresses a critical challenge in network analysis: how to release community labels from spectral clustering without leaking sensitive relational information. The method uses a stability test to release a noisy empirical spectral embedding, with perturbation bounds providing computable stability certificates and local sensitivity. For ordinary networks, it achieves edge-level differential privacy (edge-DP) and, under the degree-corrected stochastic blockmodel, separates the non-private clustering error from the privacy-induced error. This guarantees weak consistency in sparse networks and exact recovery in moderately sparse networks, with a matching lower bound proving the privacy budget is optimal up to logarithmic factors.
For bipartite networks, NetPTR extends to column-node-DP, protecting an entire right-side incidence profile while inferring left-side community structure. Consistency is proven under a bipartite degree-corrected block model. Simulations and real-world examples demonstrate the privacy–accuracy tradeoff. With 35 pages, 5 figures, and 2 tables, the paper is a comprehensive contribution to privacy-preserving network analysis, offering both theoretical guarantees and practical algorithms for sensitive social and information networks.
- NetPTR uses a stability test to release noisy spectral embeddings with edge-DP for ordinary networks and column-node-DP for bipartite networks.
- Achieves weak consistency in sparse networks and exact recovery in moderate sparse networks under the degree-corrected stochastic blockmodel.
- A matching lower bound proves the required privacy budget is sharp up to logarithmic factors, ensuring optimal privacy-accuracy tradeoff.
Why It Matters
Allows privacy-preserving community detection in social and information networks without sacrificing accuracy.