17.2.3 Multidimensional Scaling
A third class of applications uses multidimensional scaling (MDS) [3] algorithms to
visualize relative large data collections. There exists a wide variety of MDS algorithms,
such as classical scaling [47] (e.g. used to visualize music playlists by Donaldson
[11]), nonmetric MDS [30], Sammon Mapping [42] (e.g. used to visualize a
image collection by Pe˘cenovi´c et al. [37]), and SMACOF [10] (e.g. used to visualize
a roller skates catalog by Stappers et al. [46]). However, they are all based on
mapping a (symmetric) dissimilarity matrix into low dimensional Euclidean space,
such that distances between pairs of points represent the dissimilarities as closely as
possible. MDS is one of the methods we use to create product catalog maps and is
better described in the next section.