Community detection in bipartite networks with stochastic block models

Area of Science:

• Network Science
• Statistical Modeling
• Computer Science

Background:

• Bipartite networks exhibit disassortative community structures.
• Stochastic block models (SBM) are flexible for network community detection.
• Existing SBMs do not fully leverage bipartite network properties.

Purpose of the Study:

• Introduce a Bayesian nonparametric SBM for bipartite networks (biSBM).
• Develop an efficient algorithm for bipartite community detection.
• Parsimoniously determine the number of communities.

Main Methods:

• Bayesian nonparametric formulation of the SBM.
• Development of an efficient algorithm for community detection.
• Comparative analysis with existing SBMs and hierarchical models.

Main Results:

• biSBM enhances community detection in noisy bipartite networks.
• Achieves a sqrt[2] improvement in model resolution limit.
• Identifies a regime where nonhierarchical models outperform hierarchical ones for smaller, sparser networks.

Conclusions:

• The biSBM offers superior performance for bipartite community detection.
• Provides insights into the optimization landscape of community detection.
• Challenges traditional views on model selection for specific network types.