What we have done at CDL in our access repository based on is
sort of like your option 3...  The system we are using was designed for
full text searching.  If an object has a TEI or an EAD that is indexed,
then the METS is mostly ignored at indexing.  If the object is something
like a photo scrapbook or a simple image with no TEI or EAD, then METS
is indexed.  The top level DMD (actually we cheat and assume the first
DMD with xmlData is for the parent object) is turned into the dublin
core metadata for the METS document as a whole; and where there is more
DMD -- that is all indexed as if its the text of the document.  This
way, a metadata+text keyword search will find hits on words in the deep
metadata and we will see "snippets" in the search results because our
search results think they are hits in the text of the textsless
document.  I may be able to get it so that clicking on the faux-text hit
in the snippet will open up the page viewer to the page where the hit is
in the metadata.

Some people have objected that this approach blurs the line between
content and metadata because metadata is in the "text" index.  I think
this is only a temporary solution, ideally you could search across
different levels of discovery granularity, and limit to specific fields
at any level. 

Also, most of our METS are at the level of the individual item / photo /
album with one descriptive record for the whole entity.  One reason we
went to METS was because people were not satisfied with searching EAD
finding aids and having to dig down the description of subordinate
components (dsc) to find the digitized item.  They wanted an item level
search of all the digital objects.  Now that we have started to accept
submissions of more complex METS, we are seeing more and more cases
where significant portions of the dsc hierarchy has been moved to the
METS structMap -- and we have the same issue of people wanting to search
/ retrieve on different levels of the hierarchy, because most of the
item level cataloging is buried in deep metadata. 

-- Brian

On Mon, 2005-11-21 at 06:46, mfeltner wrote:
> I am working on a small cultural heritage collection that features a few 
> scrapbooks and photo albums from the early to mid-1900s.  This collection is 
> the first at our institution to utilize METS for complex objects.  Given how 
> new we are to METS, we're still feeling out how to make best use of it -- as 
> well as cope with the limitations of our digital content management software.
> For each scrapbook/album, I am creating METS records featuring two levels of 
> descriptive metadata: (1) a parent DMD for the object as a whole; and (2)  
> child DMDs for the many individual photos/drawings on the pages.  Our grant 
> project is particularly fortunate to have a historian on board, which has 
> allowed us to create rich descriptive records for most individual photos in 
> albums and scrapbooks.  Perhaps the most important feature of these records is 
> the identification of people in photos.  These names are obviously best 
> captured in the child DMD for each photo, rather than the parent DMD.
> I am curious how others working with similar materials are utilizing the many 
> descriptive metadata records within a single METS file.  I would like to see 
> these records exploited to their fullest capacity for search and discovery, 
> but am unsure what would be the best scenario to make that happen.  Our system 
> breaks METS objects into their many component objects.  What this means for 
> resource discovery is that child objects as well as parent METS objects are 
> searched and retrieved.  So a search that matches a child DMD will retrieve 
> that component image file and child DMD, as well as the entire METS object and 
> parent DMD.  For those of you dealing with complex, image-based materials like 
> albums and scrapbooks, how are you allowing your many DMDs to be searched and 
> retrieved?
> Given our specific software in mind, it looks as if our collection may have at 
> least three options:
> 1. Allow only parent DMDs to be searched/retrieved through resource discovery, 
> but allow child DMDs to be viewed as the user pages through the METS object as 
> a whole.
> This kind of functionality might be possible if we can deactivate 
> search/retrieval of child DMDs in our software.  According to this scenario, 
> the child DMDs would *not* function as *access* points, but could provide 
> additional information if a user finds a particular photo/drawing for which 
> he/she would like more detail.  One particular problem this raises is the 
> inability/difficulty of finding photos of specific people that are located in 
> albums/scrapbooks through the search interface.  For example, if one searches 
> Roosevelt and a scrapbook contains a picture of Roosevelt, but that name is 
> only captured in a child DMD, resource discovery will not retrieve that image 
> or scrapbook.
> 2. Allow both parent and child DMDs (and corresponding objects) to be searched 
> and retrieved.
> This is the current functionality supported by our software.  Using the 
> previous example of searching Roosevelt, this would result in both the 
> specific image of Roosevelt being retrieved (with this record indicating that 
> this child object is part of a particular scrapbook), as well as the scrapbook 
> as a whole.  Even if the relationship to the parent is specified in the child 
> DMD, do you think this could be confusing for users?
> 3. Allow both parent and child DMDs to be searched, but retrieve only the 
> parent METS object.
> Actually, I'm not even sure if this is possible in our software, but we can 
> always ask for enhancements, right?
> Using the Roosevelt example again, this would result in the full scrapbook 
> being retrieved for this query.  The parent DMD for the scrapbook, however, 
> mentions nothing of Roosevelt, so this might result in confusion/frustration 
> for the user.  They might interpret this as a false hit or otherwise get tired 
> of paging through the scrapbook looking for a needle in a haystack, as it 
> were.
> Unfortunately, our software does not include functionality that would allow 
> the scrapbook to be retrieved but opened to the particular page on which 
> Roosevelt is pictured.  This, to me, would be the best option, as access to 
> the individual item would be preserved, but the item would also never be 
> viewed outside its original context within the scrapbook.
> Any comments/feedback on these options would be greatly appreciated.  Do any 
> of these three sound better/worse than the others?  Can anyone think of 
> alternative scenarios that would better utilize our metadata and facilitate 
> access to important pieces of a whole?
> Many thanks,
> Melanie
> --------------
> Melanie Feltner-Reichert
> Digital Coordinator
> IMLS Funded Digital Collection:
> "From Pi Beta Phi to Arrowmont"
> John C. Hodges Library
> University of Tennessee
> Knoxville, Tennessee
> Email: [log in to unmask]