A distributed GPU-based Correlation Clustering algorithm for large-scale signed social networks - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2017

A distributed GPU-based Correlation Clustering algorithm for large-scale signed social networks

Un algorithme de correlation clustering basé sur GPU et distribué pour les réseaux sociaux signés à grande échelle

Lúcia Drummond
  • Fonction : Auteur
Rosa Figueiredo
Yuri Y. Frota
  • Fonction : Auteur
  • PersonId : 988141

Résumé

When applied to signed networks, the Correlation Clustering (CC) problem consists of an important tool to study how balanced a social group behaves and if this group might evolve to a possible balanced state. Solving such combinatorial optimization problem is a challenging task, which heavily relies on heuristic procedures, one of the few solution methods capable of analyzing large network instances. In this work, we present a scalable method to solve Correlation Clustering (CC) problems on large-scale signed networks. A distributed GPU-powered version of the ILS metaheuristic, which benefits from data parallelism, has been developed. This approach is horizontally scalable and efficient, as it provides good quality clustering results when compared to non-distributed methods. Experiments were conducted on both synthetic and real datasets. The proposed algorithm achieved improved solution values when compared to the existing parallel solution method. In particular, one of the largest graphs we have considered in our experiments contains 1 million nodes and 8 million edges-such graph can be clustered in two hours using our algorithm. The proposed method can process networks for which there is no efficient solution using the existing algorithms found in the literature.
Fichier principal
Vignette du fichier
artigo-ils-distribuido-wscad.pdf (578.18 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-02987509 , version 1 (03-11-2020)

Identifiants

  • HAL Id : hal-02987509 , version 1

Citer

Mario Levorato, Lúcia Drummond, Rosa Figueiredo, Yuri Y. Frota. A distributed GPU-based Correlation Clustering algorithm for large-scale signed social networks. Brazilian Symposium on High Performance Computing Systems, Oct 2017, Campinas, Brazil. ⟨hal-02987509⟩
9 Consultations
95 Téléchargements

Partager

Gmail Facebook X LinkedIn More