Re: [xml-dev] Future of Databases

> A central issue of any querying system is to define or discover
> the appropriate vocabulary for that domain.

Couldn't find the link to the Web Techniques article about this. It
described two search strategies -- domain vocabularies and using machine
intelligence to recognize patterns in data that permit similarity searches.
The former has been done at the National Library of Medicine since the '60s.
The latter was implemented in the Whirl search engine developed by William
Cohen at AT&T.

My suggestion was combining both approaches  -- use domain experts to keep
the vocabualry current but supplement their efforts with machine processing.
Use computers to identify the emergence of new terms and concepts in the
literature.

> Which constraint system forms the basis for the query? For example,
querying
> a video for frame identifiers is straightforward.

One of the reasons database vendors moved to object-relational technology
was to provide database plug-ins for adapting a database to application
requirements. They are doing it now for XML but they've previously done it
for images and video to support content-based queries.

The classic example for Informix is the sunset query implemented for a
database of the California Department of Water Resources. About 40,000
slides were captioned, digitized, and geo-coded for the database. The
developers added a sunset function for querying the images by doing a color
match for orange at the top of the photos. That enables someone, for
example, to find all sunset photos taken within 20 miles of Monterrey. DB2
also supports content-based queries. Query by Image Content can match on
colors, textures, etc.

Those capabilities have been around for several years. The ability to query
rich types is one of the reasons I've advocated using databases to house XML
collections. You're not limited to doing queries over text documents when
your application might need maps, images, video or other types.