Sorry to jump in late on this...

We've been experimenting with a hybrid XML/SQL approach for our digital
objects at Brown.

We are constructing our metadata as pure XML structures (METS) and then
populating a repository database (MySQL) using the MODS components from
these original records.   Our MODS database contains over 20 tables -- most
of which correspond to higher level MODS nodes such as title and
origininfo.  Additional tables record the object level data and associate
an object with a "catalog view" stylesheet, owners, and component files
(for management purposes).  Once we have loaded data into the tables (part
of our SIP procedure) we retain the METS record along-side the
database.  The METS records are needed later in this model.  The only
direction of interaction is XML to SQL -- we have no need (or desire!) to
recreate a METS record from these tables.

We next create two inverted indexes from our MODS tables -- one for phrases
and one for terms.  Both index tables have associated mapping tables to
link terms/phrases to records and fields within records.  The mappings will
tell us, for example, if a term or phrase appears in a main title,
alternative title, uniform title, or a title within a constituent item, so
we are confident that our searching can be made relatively precise.

We are using (for now) PHP to manage database interactions and to handle
the arrays needed to accommodate multi-term searching in a manner more
efficient that having MySQL perform a series of joins and
sub-selects.  Once an appropriate record is found in this database
architecture, the METS record takes over and drives the remainder of the
user's interaction with the digital object.  The objects table provides the
linking mechanism between an object and its top-level XSLT viewer.

As I've alluded to several times, this is an experiment, and we're not sure
if we'll ultimately go with this model or seek out a purely XML based
solution.  Since the availability of lost-cost effective XML search tools
is still somewhat scant, the hybrid approach does have some appeal.

Patrick



Patrick Yott
Center for Digital Initiatives
Brown University Library
(401) 863-2817 (office)
(401) 863-1272 (fax)