ASAClu: selecionando clusters diversos e relevantes

Abstract

No clustering algorithm is guaranteed to find actual groups in any dataset. To deal with this problem, many techniques apply various clustering algorithms to a dataset, generating a set of partitions and assessing them to select the most appropriated ones. The problem in selecting partitions is that redundancy can be seen inside partitions, as the same cluster can appear in different partitions. Also, one can underestimate the quality of a cluster, assessing only the quality of a partition. For these reasons, a new selection strategy named ASAClu is aimed at selecting a relevant and diverse subset of clusters instead of partitions, given an initial collection.