Social Network Analysis

Statistical methods for community detection, multi-layer networks, network covariates, and longitudinal network modeling

Overview

This page highlights several of my contributions to statistical network analysis, spanning community detection, multi-layer and hypergraph modeling, network covariate adjustment, and inference for evolving (longitudinal) networks.

Social Network Analysis

Across these projects, the common themes include:

  • Scalable statistical modeling for large real-world networks
  • Integration of covariates and nodal attributes into community and structure inference
  • Higher-order network analysis (hypergraphs, multi-layer networks, co-evolving systems)
  • Formal model selection tools for determining the number of communities
  • Inference frameworks that balance statistical rigor and computational feasibility

My work connects classical network modeling with modern statistical learning, offering both theoretical guarantees and practical algorithms.

(He et al., 2025)

📈 Semiparametric Modeling and Analysis for Longitudinal Network Data

Yinqiu He, Jiajin Sun, Yuang Tian, Zhiliang Ying, & Yang Feng, Annals of Statistics, 2025
📄 Paper | 💻 Python Package: DLSM

Highlights

  • Proposes a flexible semiparametric framework for dynamic networks observed over time
  • Separates time-varying nonparametric components (e.g., evolving interaction patterns) from parametric effects
  • Allows network dependence, temporal smoothness, and heterogeneous evolution
  • Offers theory for estimation, inference, and model identifiability
  • Applicable to longitudinal relational data in public health, social sciences, neuroscience, and more

🧬 Spectral Clustering via Adaptive Layer Aggregation for Multi-Layer Networks

Sihan Huang, Haolei Weng, & Yang Feng, Journal of Computational and Graphical Statistics, 2023
📄 Paper | 💻 Python Package: multi_net

Highlights

  • Tackles multi-layer networks, where each layer may represent a different relation
  • Introduces an adaptive aggregation strategy that
    • weighs layers by signal quality
    • reduces noise contamination
    • enhances community detection accuracy
  • The method is computationally efficient and scalable
  • Includes consistency theory and finite-sample guarantees
  • Useful for biological networks, multi-platform social networks, multi-omics integration

🧩 PCABM: Pairwise Covariates-Adjusted Block Model

Sihan Huang, Jiajin Sun, & Yang Feng, Journal of the American Statistical Association, 2023
📄 Paper | 💻 Python Package: PCABM

Highlights

  • Extends the classical stochastic block model (SBM) by incorporating pairwise covariates
  • Separates community structure from observed relationships, reducing confounding
  • Supports likelihood-based estimation and community recovery
  • Improves interpretability by adjusting for user similarity, geographic distance, demographics, etc.
  • Widely applicable to social networks, recommendation systems, relational public health data

🧠 Community Detection with Nodal Information

Haolei Weng & Yang Feng, Stat, 2022
📄 Paper 

Highlights

  • Integrates node-level covariates into likelihood and variational inference frameworks
  • Combines network topology with attribute information for improved clustering
  • Variational approximation enables fast computation for large networks
  • Particularly effective when network structure alone is weak or ambiguous
  • Examples: political networks, brain networks, citation networks

🔺 Testing Community Structure for Hypergraphs

Mingao Yuan, Ruiqi Liu, Yang Feng, & Zuofeng Shang, Annals of Statistics, 2022
📄 Paper

Highlights

  • Develops statistical tests for community structure in hypergraphs
  • Handles higher-order interactions (e.g., groups of 3+, team-based interactions)
  • Provides minimax testing boundaries
  • Bridges traditional graph theory with complex k-way interactions
  • Applications in group behavior analysis, collaborative teams, co-authorship networks

❓ How Many Communities Are There?

Diego Franco Saldana, Yi Yu, & Yang Feng, Journal of Computational and Graphical Statistics, 2017
📄 Paper 

Highlights

  • Provides formal tools for estimating the number of communities
  • Combines spectral properties with statistical model selection criteria
  • Works for SBMs, DCBMs, and their generalized variants
  • A foundational method used across statistical network modeling
  • Useful for practitioners selecting model complexity in applications

If you would like to learn more about my work or explore code, slides, and related materials, please check my full list of publications on the Publications page or contact me.