We consider the problem of partitioning, in a highly accurate and highly e±cient way, a set of n documents lying in a met- ric space into k non-overlapping clusters. We augment the well-known furthest-point-¯rst algorithm for k-center clus- tering in metric spaces with a ¯ltering scheme based on the triangular inequality. We apply this algorithm to Web snip- pet clustering, comparing it against strong baselines consist- ing of recent, fast variants of the classical k-means iterative algorithm. Our main conclusion is that our method attains solutions of better or comparable accuracy, and does this within a fraction of the time required by the baselines. Our algorithm is thus valuable when, as in Web snippet clus- tering, either the real-time nature of the task or the large amount of data make the poorly scalable, traditional clus- tering methods unsuitable.
A scalable algorithm for high-quality clustering of Web snippets
Geraci F;Pellegrini M;Sebastiani F
2006
Abstract
We consider the problem of partitioning, in a highly accurate and highly e±cient way, a set of n documents lying in a met- ric space into k non-overlapping clusters. We augment the well-known furthest-point-¯rst algorithm for k-center clus- tering in metric spaces with a ¯ltering scheme based on the triangular inequality. We apply this algorithm to Web snip- pet clustering, comparing it against strong baselines consist- ing of recent, fast variants of the classical k-means iterative algorithm. Our main conclusion is that our method attains solutions of better or comparable accuracy, and does this within a fraction of the time required by the baselines. Our algorithm is thus valuable when, as in Web snippet clus- tering, either the real-time nature of the task or the large amount of data make the poorly scalable, traditional clus- tering methods unsuitable.File | Dimensione | Formato | |
---|---|---|---|
prod_172145-doc_79653.pdf
solo utenti autorizzati
Descrizione: A scalable algorithm for high-quality clustering of Web snippets
Tipologia:
Versione Editoriale (PDF)
Dimensione
168.35 kB
Formato
Adobe PDF
|
168.35 kB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.