Hybrid clustering for large sequential data

Abstract

Many scientific and commercial domains have witnessed enormous data explosion that has inherent sequential nature. While clustering sequential data is useful for various purposes, there has been less success due to the discrete nature of sequential data. We combine techniques from data mining and computational geometry to efficiently and effectively segment sequential web usage data in data-rich environments. We provide an hybrid O(n n) clustering algorithm that combines medoid-based partitioning and agglomerative hierarchial clustering. This hybridization is inspired by geometrical and topological aspects of the Voronoi diagram. Experimental results demonstrate the superiority of our hybridization over traditional approaches.