Multi-set spectral clustering of time-evolving networks using the supra-Laplacian

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Froyland, Gary, Kalia, Manu, Koltai, Péter
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866914051579183104
author Froyland, Gary
Kalia, Manu
Koltai, Péter
author_facet Froyland, Gary
Kalia, Manu
Koltai, Péter
contents Complex time-varying networks are prominent models for a wide variety of spatiotemporal phenomena. The functioning of networks depends crucially on their connectivity, yet reliable techniques for learning communities in time-evolving networks remain elusive. We adapt successful spectral techniques from continuous-time dynamics on manifolds to the graph setting to fill this gap. We consider the supra-Laplacian for graphs and develop a spectral theory to underpin the corresponding algorithmic realisations. We develop spectral clustering approaches for both multiplex and non-multiplex networks, based on the eigenvectors of the supra-Laplacian and specialised Sparse EigenBasis Approximation (SEBA) post-processing of these eigenvectors. We demonstrate that our approach can outperform the Leiden algorithm applied both in spacetime and layer-by-layer, and we analyse voting data from the US senate (where senators come and go as congresses evolve) to quantify increasing polarisation in time.
format Preprint
id arxiv_https___arxiv_org_abs_2409_11984
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Multi-set spectral clustering of time-evolving networks using the supra-Laplacian
Froyland, Gary
Kalia, Manu
Koltai, Péter
Social and Information Networks
Dynamical Systems
Physics and Society
Complex time-varying networks are prominent models for a wide variety of spatiotemporal phenomena. The functioning of networks depends crucially on their connectivity, yet reliable techniques for learning communities in time-evolving networks remain elusive. We adapt successful spectral techniques from continuous-time dynamics on manifolds to the graph setting to fill this gap. We consider the supra-Laplacian for graphs and develop a spectral theory to underpin the corresponding algorithmic realisations. We develop spectral clustering approaches for both multiplex and non-multiplex networks, based on the eigenvectors of the supra-Laplacian and specialised Sparse EigenBasis Approximation (SEBA) post-processing of these eigenvectors. We demonstrate that our approach can outperform the Leiden algorithm applied both in spacetime and layer-by-layer, and we analyse voting data from the US senate (where senators come and go as congresses evolve) to quantify increasing polarisation in time.
title Multi-set spectral clustering of time-evolving networks using the supra-Laplacian
topic Social and Information Networks
Dynamical Systems
Physics and Society
url https://arxiv.org/abs/2409.11984