A Comparison of Experiments with the Bisecting-Spherical K-Means Clustering and SVD Algorithms

Jiménez Gonzalez, Daniel     dajigon@doctor.upv.es     (DSIC Department Polytechnic University of Valencia, Departamento de Sistemas Informáticos y Computación)   Enguix, Carlos F.     carlos@must-es.com     (Mediterranean University of Science and Technology, Campus Universidad Politécnica de Valencia)   Vidal Gimeno, Vicente     vvidal@dsic.upv.es     (DSIC Department Polytechnic University of Valencia, Departamento de Sistemas Informáticos y Computación)

Jornadas de Tratamiento y Recuperación de la Información / Departamento de Biblioteconomía y Documentación y Departamento de Informática de la Universidad Carlos III de Madrid
(ReLIS:doi:jotris:y:2003:p42-42)

Abstract:

Several experiments comparing clustering and SVD techniques can be found in the literature. In this paper we propose a modified version of the Spherical k -means clustering algorithm, the Bisecting-Spherical k -means clustering algorithm. The bisecting clustering algorithm is used to determine the initial set in the Spherical k-means clustering algorithm. We have set up a set of experiments to compare the SVD with different number of singular values in order to find an optimal solution. Analogously we have done with our modified version of the Spherical k -means clustering algorithm, with different number of clusters. Finally we have compared both techniques with respect to precision-recall ratios, concluding that both techniques perform similarly in terms of precision-recall ratios.

Creation: 2003

 (go top)