Boucherie et al., "Adaptive cut reveals multiscale complexity in networks" (2026)

Louis Boucherie, Yong-Yeol Ahn, and Sune Lehmann, PNAS Nexus, pgag344 (2026)
DOI | arXiv

@article{boucherie2026adaptive,
    author = {Louis Boucherie and Yong-Yeol Ahn and Sune Lehmann},
    title = {Adaptive cut reveals multiscale complexity in networks},
    journal = {PNAS Nexus},
    pages = {pgag344},
    doi = {10.1093/pnasnexus/pgag344},
    year = {2026},
    archivePrefix = {arXiv},
    eprint = {2512.08741},
    primaryClass = {physics.soc-ph},
}

A common approach to identify clusters in hierarchical clustering is to cut dendrograms at a single threshold, but this is often suboptimal—especially when the dendrogram is unbalanced. We introduce the adaptive cut, which uses multi-level cuts through dendrograms via MCMC with simulated annealing. We also present a balancedness score, an information-theoretic metric for evaluating dendrogram balance. Testing across 200+ real-world networks and synthetic datasets, the approach yields significant improvements in partition density and modularity compared to traditional single-cut methods.