New algorithm for size-constrained community detection in networks

Community detection algorithms often ignore domain experts' prior expectations about community sizes. Multi-resolution methods allow indirect control of average size via a resolution parameter, but cannot enforce size limits on individual communities. This paper by Silva et al. tackles that gap by formulating both a heuristic and an exact integer optimization model for modularity optimization under strict per-community size constraints. The heuristic is designed for scalability, while the exact model serves as a baseline to validate reliability.

Using synthetic benchmarks and real-world networks, the authors demonstrate that their proposed methods avoid the pitfalls of resolution-based approaches—such as producing communities that are too large or too small for domain needs. The heuristic achieves near-optimal results while being computationally feasible for large graphs. Crucially, the implementation is publicly available in the widely-used Python Leiden algorithm package, making it accessible for practitioners in social network analysis, biology, and information science.