Partitional vs Hierarchical Clustering Using a Minimum Grammar Complexity Approach

摘要

This paper addresses the problem of structural clustering of string patterns. Adopting the grammar formalism for representing both individual sequences and sets of patterns, a partitional clustering algorithm is proposed. The performance of the new algorithm, taking as reference the corresponding hierarchical version, is analyzed in terms of computational complexity and data partitioning results. The new algorithm introduces great improvements in terms of computational efficiency, as demonstrated by theoretical analysis. Unlike the hierarchical approach, clustering results are dependent on the order of patterns' presentation, which may lead to performance degradation. This effect, however, is overcome by adopting a resampling technique. Empirical evaluation of the methods is performed through application examples, by matching clusters between pairs of partitions and determining an index of clusters agreement.