• NL-PCA provides a joint space of object and attribute category points, called
a biplot [15], while other visualization methods only provide the object points.
The category points can be used to provide an easier interpretation of the map
and to navigate through the map by selecting subsets of products.
• For categorical attributes, the category points are located in the centroids of
the object points belonging to that category. This implies that when a certain
attribute is well represented in the map, the object points are clustered around
their corresponding category value of that attribute and a selection of all points
belonging to this category will lead to a selection of a subspace of the map. For
ordinal attributes, the category point will be the point on the line closest to the
centroid of the object points.
• The third advantage is shared by NL-PCA and most other visualization methods.
That is, the distance between object points in the map is, in general, related
to dissimilarity between objects.
In Section 17.6, we also show an application of this approach to the MP3 player
data set.