Multi-dimensional scaling (MDS) creates a two (usually) dimensioanl represention of data that preserves closeness. Its input is a similarity matrix (Sik) or a dissimilarity matrix (Dik) between items an then allocates for each item i coordinates (xi,yi) such that, on average, when the Euclidean distance between (xi,yi) and (xj,yj) is small then Sik) is large (or Dik) is small) and vice versa.
Used on page 160
Also known as MDS