A new micro-objects-based evaluation measure for co-clustering algorithms

In this work, we present MOCICE-BCubed F1, a new external measure for evaluating co-clusterings, in the scenario where gold standard annotations are available for both the object clusters and the associated feature subspaces. Our proposal is an extension, using the so-called micro-objects transformation, of CICE-BCubed F1, an evaluation measure for traditional clusterings that has been proven to satisfy the most comprehensive set of meta-evaluation conditions for that task. Additionally, the proposed measure adequately handles the occurrence of overlapping in both the object and feature spaces. We prove that MOCICE-BCubed F1 satisfies the most comprehensive set of meta-evaluation conditions so far enunciated for co-clusterings. Moreover, when used for evaluating traditional clusterings, which are viewed as a particular case of co-clusterings, the proposed measure also satisfies the most comprehensive set of meta-evaluation conditions so far enunciated for the traditional task.