[Sigia-l] K Mel Faceted Classification

[Richard Hill]

[Reposted for Meltsner, Kenneth" <Kenneth.Meltsner at ca.com> to avoid 
HTML.  Dick Hill]

Facets need to be orthogonal (and hierarchical?)

The other big distinction between faceted classification and traditional 
"query-by-example" search is that the facets should be as independent or 
orthogonal to each other as possible.

If I were classifying insurance-related news, I might choose:

*  State

*  Line of business

as two (relatively) independent facets.
Orthogonality allows you to specify values (or ranges of values) for 
multiple facets with a reasonable expectation that there will be some 
results.  If the categories aren't independent, then it's too easy to 
specify searches with no results, I think.

The other issues with facets is that they need to store more than simple, 
single values.  As the facets become more numerous, the more likely it is 
that they will be quite sparse (that for many items, they will not have any 
values).  You need to ensure that facets have values most of the time 
without limiting your ability to describe very specific cases.  The 
simplest way to do this involves hierarchical facet values, I think -- an 
attribute value is broadened by going up the tree.

For example, a genre facet for novels might have values organized as a 
hierarchy:

*  Genre
-- Science Fiction and Fantasy
    -- Space opera
    -- New wave
    -- Near Future
    -- Historical fantasy
    -- Alternate History (SF)
    -- Alternate History (fantasy)
-- Mystery
    -- Police procedural
    -- Cozy
    -- Hardboiled

This probably works, although exact classification is always a 
problem:  how do you find a book like _Point of Dreams_, which is an 
alternative history (fantasy) with strong mystery (police procedural) 
elements?  Or a SF near-future hardboiled novel?  Within in a facet, it 
appears that you need to be able to handle multiple values as well, just as 
it's possible that a California insurance news item would apply to both 
auto and home insurance.

Ken Meltsner
Senior Architect, Computer Associates

Executive Director
American Society for Information Science and Technology
1320 Fenwick Lane, Suite 510
Silver Spring, MD  20910
FAX: (301) 495-0810
PHONE: (301) 495-0900

http://www.asis.org