Efficient clustering for orders

Abstract

Lists of ordered objects are widely used as representational forms. Such ordered objects include Web search results or best-seller lists. Clustering is a useful data analysis technique for grouping mutually similar objects. To cluster orders, hierarchical clustering methods have been used together with dissimilarities defined between pairs of orders. However, hierarchical clustering methods cannot be applied to large-scale data due to their computational cost in terms of the number of orders. To avoid this problem, we developed an k-o'means algorithm. This algorithm successfully extracted grouping structures in orders, and was computationally efficient with respect to the number of orders. However, it was not efficient in cases where there are too many possible objects yet. We therefore propose a new method (k-o'means-EBC), grounded on a theory of order statistics. We further propose several techniques to analyze acquired clusters of orders. © 2009 Springer-Verlag Berlin Heidelberg.