A parallel algorithm for incremental compact clustering

Abstract

In this paper we propose a new parallel clustering algorithm based on the incremental construction of the compact sets of a collection of objects. This parallel algorithm is portable to different parallel architectures and it uses the MPI library for message-passing. We also include experimental results on a cluster of personal computers, using synthetic data generated randomly and collections of documents. Our algorithm balances the load among the processors and tries to minimize the communications. The experimental results show that the parallel algorithm clearly improves its sequential version with large sets of data. © Springer-Verlag 2003.