Beyond Partitions: Allowing Overlapping Groups in Pairwise Clustering