Extending doubly stochastic scaling to bipartite graphs
Philip Knight  1@  , Luce Le Gorrec  1@  , Sandrine Mouysset  2@  , Daniel Ruiz  3@  
1 : Department of Mathematics and Statistics [Univ Strathclyde]
2 : IRIT, Université de Toulouse
IRIT
3 : Institut National Polytechnique (Toulouse)  (Toulouse INP)
Université Fédérale Toulouse Midi-Pyrénées

Doubly stochastic scaling has a long history of being used to scale a graph adjacency matrix A. It has proven useful for graph analysis. The conditions for the scaling to exist are well known and rely on the graph structure. How this extends when the (potentially rectangular) matrix A represents a bipartite graph is non straightforward though, however it would have interesting applications, e.g. for clustering bipartite graphs (co-clustering). In this work we investigate a new type of scaling that naturally enlarges doubly stochastic scaling to rectangular matrices.


Personnes connectées : 3 Vie privée
Chargement...