A variant of the spectral clustering method using l1 -penalty to promote sparse eigenvectors basis

Mélanie Blazère (1)

1 IMT - Institut de Mathématiques de Toulouse UMR5219

The spectral clustering method is one of the most well-known graph clustering algorithm, that is used in a wide range of fields and applications. In this paper, we propose a variant of the spectral clustering algorithm, called 1-spectral clustering. This procedure does not require the use of k-means to cluster the nodes of the graph, but directly estimates the indicators of the communities by computing a specific eigen-basis using a l1-penalty.

Mots clés: Spectral clustering; eigenvectors basis; community detection; l1 -penalty

Domaines: Théorie [stat.TH]

Mélanie Blazère. A variant of the spectral clustering method using l1 -penalty to promote sparse eigenvectors basis. 2016. ⟨hal-01294831⟩