Your browser doesn't support javascript.
loading
Finite-state parameter space maps for pruning partitions in modularity-based community detection.
Gibson, Ryan A; Mucha, Peter J.
Afiliación
  • Gibson RA; Department of Mathematics, University of North Carolina, Chapel Hill, NC, 27599-3250, USA.
  • Mucha PJ; Department of Computer Science, University of North Carolina, Chapel Hill, NC, 27599-3175, USA.
Sci Rep ; 12(1): 15928, 2022 09 23.
Article en En | MEDLINE | ID: mdl-36151268
Partitioning networks into communities of densely connected nodes is an important tool used widely across different applications, with numerous methods and software packages available for community detection. Modularity-based methods require parameters to be selected (or assume defaults) to control the resolution and, in multilayer networks, interlayer coupling. Meanwhile, most useful algorithms are heuristics yielding different near-optimal results upon repeated runs (even at the same parameters). To address these difficulties, we combine recent developments into a simple-to-use framework for pruning a set of partitions to a subset that are self-consistent by an equivalence with the objective function for inference of a degree-corrected planted partition stochastic block model (SBM). Importantly, this combined framework reduces some of the problems associated with the stochasticity that is inherent in the use of heuristics for optimizing modularity. In our examples, the pruning typically highlights only a small number of partitions that are fixed points of the corresponding map on the set of somewhere-optimal partitions in the parameter space. We also derive resolution parameter upper bounds for fitting a constrained SBM of K blocks and demonstrate that these bounds hold in practice, further guiding parameter space regions to consider. With publicly available code ( http://github.com/ragibson/ModularityPruning ), our pruning procedure provides a new baseline for using modularity-based community detection in practice.
Asunto(s)

Texto completo: 1 Colección: 01-internacional Base de datos: MEDLINE Asunto principal: Algoritmos / Heurística Tipo de estudio: Diagnostic_studies Idioma: En Revista: Sci Rep Año: 2022 Tipo del documento: Article País de afiliación: Estados Unidos Pais de publicación: Reino Unido

Texto completo: 1 Colección: 01-internacional Base de datos: MEDLINE Asunto principal: Algoritmos / Heurística Tipo de estudio: Diagnostic_studies Idioma: En Revista: Sci Rep Año: 2022 Tipo del documento: Article País de afiliación: Estados Unidos Pais de publicación: Reino Unido