AntClust: Ant Clustering and Web Usage Mining

Abstract. In this paper, we propose a new ant-based clustering algorithm called AntClust. It is inspired from the chemical recognition system of ants. In this system, the continuous interactions between the nestmates generate a “Gestalt” colonial odor. Similarly, our clustering algorithm associates an object of the data set to the odor of an ant and then simulates meetings between ants. At the end, artificial ants that share a similar odor are grouped in the same nest, which provides the expected partition. We compare AntClust to the K-Means method and to the AntClass algorithm. We present new results on artificial and real data sets. We show that AntClust performs well and can extract meaningful knowledge from real Web sessions.

• Type: Conference
• Year: 2003
• Where: GECCO
• Authors: Nicolas Labroche, Nicolas Monmarché, Gilles Venturini
• Keywords: Ant-based Clustering Algorithm, Clustering Algorithm, Data Sets, GECCO 2003