Email list hosting service & mailing list manager


Re: Multi-dimensional Indexing Gerry Mckiernan 19 Jan 2001 17:37 UTC

      In response to my recent posting concerning the use of
'Multi-Dimensional Indexing' in e-Books or e-Journals

[  http://sunsite.berkeley.edu/Web4Lib/archive/0101/0180.html ] ,

Peter B. Boyce, Senior Consultant for Electronic Publishing for the
American Astronomical Society (AAS) sent the response below [Re-posted
with permission]

   He sites/cites a most impressive application of the Kohonen
self-organizing maps for an index to the _Astrophysical Journal_ If there
is one special Web site you visit today, I strongly recommend that you
visit and explore this most remarkable Multi-Dimensional Index'

/Gerry McKiernan
Self-Organized Librarian
Iowa State University
Ames IA 50011

gerrymck@iastate.edu

P.S. I have been a Big Fan of the Kohonen SOM for several years. For other
examples, please see _The Big Picture(sm)_ [
http://www.public.iastate.edu/~CYBERSTACKS/BigPic.htm] my clearinghouse
devoted to "Visual Browsing in Web and non-Web Databases", notable the two
examples under Helsinki University of Technology, the home of Teuvo
Kohonen, the developer of the SOM

>>> "Peter B. Boyce" <pboyce@aas.org> 01/18/01 02:34PM >>>
Gerry,

The Centre de Donnees Astronomique de Strasbourg
(http://cdsweb.u-strasbg.fr/) utilizes the technique of "Self-organizing
maps" to collect articles under various concept headings and arranging
those headings such that adjacent concepts are closely related to each
other. The entire collection is then displayed as a three dimensional map.
See http://simbad.u-strasbg.fr/ApJ/map.pl for an example of how this
works.

This provides a visual method of navigating among concepts and discovering
related articles in a visual way which is much more intuitive to me than
text-based searches -- even when they are incorporated into a "concept
map" of the type described by Ross.

They use a neural network method which includes all the article keywords
to characterize each article. The article keywords -- never less than
three per article -- are assigned by the editor and the vocabulary is
controlled, so this method is quite reliable. Unfortunately, this neural
network approach uses a lot of computing power, and it is not suitable for
"on the fly" usage.  But, it demonstrates the kind of approach which will
have to be done eventually so that readers can find relevant information
more reliably than at present.

Cheers,
--Peter--
_________________________________________________________
Peter B. Boyce    -   Senior Consultant for Electronic Publishing, AAS
email: pboyce@aas.org
Summer address:                                Winter: 4109 Emery Place,
33 York St., Nantucket, MA 02554        Washington, DC 20016
Phone:  508-228-9062                           202-244-2473
_________________________________________________________