Core Concepts
GSL-LPA introduces a parallel implementation of LPA to address internally-disconnected communities, outperforming existing methods significantly.
Abstract
GSL-LPA addresses the issue of internally-disconnected communities in community detection algorithms. It surpasses FLPA, igraph LPA, and NetworKit LPA by 55Γ, 10, 300Γ, and 5.8Γ respectively. The algorithm achieves a processing rate of 844π edges/s on a graph with 3.8π΅ edges.
The content discusses the problem of community detection and the importance of efficient parallel algorithms like GSL-LPA. It introduces GSL-LPA as a solution to internally-disconnected communities identified by other algorithms like FLPA, igraph LPA, and NetworKit LPA.
The article explains the Label Propagation Algorithm (LPA) and its susceptibility to identifying internally disconnected communities. It presents GSL-LPA as an improved version that resolves this issue efficiently.
Key metrics used to support the argument include processing rates achieved by GSL-LPA compared to other algorithms like FLPA, igraph LPA, and NetworKit LPA.
Stats
Our experiments show that GSL-LPA outperforms FLPA by 55Γ.
On a graph with 3.8π΅ edges, GSL-LPA achieves a processing rate of 844π edges/s.
GSL-LPA surpasses igraph LPA by 10 times.
NetworKit LPA is exceeded by GSL-LAP by 5.8 times.
Quotes
"GSL-LAP not only mitigates this issue but also surpasses FLAP, igraph LAP, and NetworKit LAP."
"GSL-LAP scales at a rate of 1.6Γ for every doubling of threads."