Fast Ant-Inspired Clustering Algorithm for Web Usage Mining
Résumé
In this paper, we propose a new agglomerative clustering algorithm named Leader Ant (LA) that improves the classical Leader clustering algorithm model [14] with a metaphor inspired by the chemical recognition system of ants. In our approach, each object of the data set is associated to the colonial odour of an artificial ant. At each iteration , a randomly chosen ant meets ants from each already existing nest to decide if it integrates this nest or if it creates its own nest. At the end of this iterative meeting process , the nests represent a partition of the intial data set. Similarily to the Leader algorithm, LA processes each data only once, which allows short computation times even on large data sets. LA is compared to other clustering algorithm such as k-means or AntClust [9] on artificial and real data sets. Finally, we briefly describe results obtained when applying LA on real Web usage data from a French museum Web site.
Origine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...