Synergy between OPOSSUM and CLUSION

The visualization and clustering techniques presented in this work need to be considered together and not in isolation. This is because CLUSION is particularly suited to viewing the output of OPOSSUM. First, the similarity matrix is already computed during the clustering step, so no extra computation is needed, except for permuting this matrix, which can be done in O(n) since the size and seriation order of each partition is known. Second, since METIS involves boundary Kernighan-Lin refinement, clusters that are similar appear closer in the seriation order. Thus it is no coincidence that clusters $\mathcal{C}_1$ and $\mathcal{C}_2$ appear contiguous in figure 3.6(a). Finally, one can experiment with different similarity measures for OPOSSUM and quickly get visual feedback regarding their effectiveness using CLUSION (figure 3.7).