Here is some pretty old work that did the same sort of thing. The self organizing map (SOM) is an interesting alternative to MDS since it allows mapping a low dimensional approximate manifold to a linear space. The basic idea is that it preserves close distances and doesn't much care about distances to far away points. Similar results should be obtainable using local linear embedding (LLE)
http://comminfo.rutgers.edu/~aspoerri/Teaching/InfoVisResources/papers/UIR-1996-01-Card-CGA-VisSurvey.pdf http://aclweb.org/anthology-new/X/X96/X96-1032.pdf http://www.tgc.com/dsstar/99/0518/100758.html On Mon, Aug 27, 2012 at 12:58 AM, Dmitriy Lyubimov <[email protected]>wrote: > Mds is usually a way to visualize it close to truth. The rest looks like > countours of a regular 2d kernel density estimate. > On Aug 26, 2012 8:02 PM, "Lance Norskog" <[email protected]> wrote: > > > This is a really cool 3D visualization of a tag cloud with distances: > > http://langtech.jrc.ec.europa.eu/Pictures/ThemeScape-overview_EP259.pdf > > > > What is the sequence to make this? I'm thinking: > > 1) Create a document/term matrix. > > 2) Random Projection of term vectors onto 2D. > > 2D distances match N-dimensional distances between terms. > > 3) Do SVD of term vectors. > > 4) Use first feature vector to select height of each term. > > Or, norm of the feature vector X singular values. > > > > After this, the mapping software does the rest of the work via topo > > and word placement algorithms. > > > > -- > > Lance Norskog > > [email protected] > > >
