Addressing the validation of overlapping clustering algorithms

Abstract

The validation of overlapping clusterings has gained considerable attention in the last years, as it represents a fundamental step towards achieving reliable clustering algorithms, still being largely considered as an open question. A number of measures have been proposed for validating overlapping clusterings, but it remains unclear which one is the most convenient. Most existing external measures aim to validate partitional clusterings, even when most real-life problem involve overlapping clustering scenarios. In this paper, we propose a new external measure specifically designed for validating overlapping clusterings, which fulfills the main set of desirable conditions presented in the literature. We also show that our proposal correctly handles situations where previous measures display undesirable performances.