a product catalog, one could use the dissimilarity measure we introduce in the next
section, that is able to handle missing values and mixed attribute types.
The treemap approach has two drawbacks. First of all, the original treemap algorithm
often produces tall rectangles as results for small clusters or single items,
which makes visualization of single products quite hard. This problem can be partly
overcome by using squarified treemaps [4] or, even better, quantum treemaps [2],
which guarantee a certain aspect ratio for the rectangles representing the items. The
second drawback is that, although similar products are clustered together, there is
no clear distance interpretation and ordering between clusters might be lost, that is,
two quite similar products that are assigned to two different clusters might not be
close to each other in the map.