An issue at hand for me is how to set up in SRW+CQL a search
on "title" at the Library of Congress.
I tried using the index "dc.title" with record schema "dc".
But, I get disconcerting results--most of the retrieved
records do not have my search term value in the XML element
called "dc.title". On the surface, this looks badly broken.
Larry Dixson explained to me:
"Our title index is pretty broad. Below are the MARC
fields and subfields that are included in the index
for a keyword title search:
100/fgklnpt,110/dfgklnpt,111/fgklnpt,130/adfghklmnoprst,
210/ab,222/ab,240/adfghklmnoprs,242/abhnp,
243/adfghklmnoprs,245/abfghknps,246/abfghnp
247/abfghnp,249/abfghnp,400/fgklnptv,410/dfgklnptv,
411/dfgklnptv,440/anpv,490/av,505/t,534/t,580/a,581/a,
600/fghklmnoprst,610/dfghklmnoprst,611/fghklnpst,
630/adfghklmnoprst,700/fghklmnoprst,710/dfghklmnoprst,
711/fghklnpst,730/adfghklmnoprst,740/ahnp,760/cgst,
762/cgst,765,767,770,772,773,774,775,776,777,780,
785,786,787 subfields cgkst,800/fghklmnoprstv,
810/dfghklmnoprstv,811/fghklnpstv,830/adfghklmnoprstv,
890/ai"
[...]
"All of those fields/subfields contain titles.
We map our intersite searches to the most generic
internal _title_ search available."
I guess that a "correct" solution here might be for LoC
to concatenate the values of all fields searched when
constructing the value of the XML element "dc.title".
Of course, that would be quite an ugly record and not
really following the spirit of the "dc" record schema.
My feeling is that the problem lies more with the abstract
access point. Given the mapping that Larry has for the
LoC MaRC fields, perhaps the index ought to be named
simply "title" rather than "dc.title". That way, there
is no expectation that the abstract search index should
be an exact match to the record schema element.
In conversation with Sebastian, it turns out that the
IndexData search actually _does_ behave that way: one
can search on "title" as though that index name were
a CQL built-in index. Very Cool!
Maybe it would be a good idea to make this approach
a recognized behavior of CQL? I can envision a short
list of several real common indexes that could be
mapped wherever at the choice of the server--sort
of like a "cql.anywhere with facets".
Eliot
|