Hi all,

I'm uploading a refresh of the open library data that includes two main
changes: linkage to the new LCSH dataset and a first crude stab at
FRBRisation.

To provide the subject linkage I downloaded the dump of RDF from
http://id.loc.gov/authorities/ and parsed it to create a local database
mapping skos:prefLabel to URI (using gdbm). Then I modified my json2rdf.py
script to use this database and look up the subjects.

An example of an item where all the subjects matched:

<http://ol.dataincubator.org/works/6102>

This worked less well:

<http://ol.dataincubator.org/works/5991>

And this didn't match any:

<http://ol.dataincubator.org/works/62514>

I haven't done any in-depth analysis of whether these are subjects that
should have matched or if they are simply not available in LCSH.

For the FRBRisation I combined two approaches. For each edtiion record I
created a Work resource and one Manifestation for each ISBN listed in the
record. I remembered that dublin core had isVersionOf/hasVersion properties
which seemed to fit the semantic I wanted so I used them to link Works
directly to Manifestations bypassing any Expression layer. The results of
this approach were moderately successful as can be seen by this work which
has 5 manifestations.

<http://ol.dataincubator.org/works/61705>

The main problem is that there is no distingushing information between the
versions. To address this I grabbed a dump of ThingISBN (which is free for
non-commercial use) and parsed it to create a lookup database of ISBN ->
work id. I think used that to link Manifestations to Works, only creating a
new work resource where there was no work id in ThingISBN.

The coverage of ThingISBN turns out to be pretty good. Here's a work with 3
Manifestations:

<http://ol.dataincubator.org/works/60798>

Since I'm only working with 1% of the total open library corpus I wanted to
increase the number of records that were likely to be related. I grepped the
entire corpus for hobbit and tolkien and ran those records through the
converter.

Here's The Hobbit, which looks pretty good:

<http://ol.dataincubator.org/works/59650>

Here's another work that represents The Hobbit:

<http://ol.dataincubator.org/works/152602>

And the annotated Hobbit:

<http://ol.dataincubator.org/works/152611>

I also made some minor formatting changes for skos:prefLabel which is now
composed of title_prefix, title and subtitle. I also removed the
dct:isPartOf properties that I added in last time - that was my
misunderstanfing of the void convention on datasets. I did a tiny bit of
work on formatting series titles but it needs a whole lot more work - I
think a lot of information has been lost by open library in their conversion
from MARC.

Ian

On May 4, 4:05 pm, Ian Davis <m...@iandavis.com> wrote:

Nice Ian. Just a couple of minor comments:

1) as w/periodicals, would be nice to have publishers as URIs if
possible; maybe http://publishers.dataincurbator.org?

2) the versions/manifestations; do you have the input data to be able
to assign them a bibo type as well?

Bruce

Thanks :)

Yes, that's a good idea. Apart from the publisher data in open library and
periodicals sets is there a good database we could target for conversion?

The records contain a format indicator which could give hints. There is a
list of values I have come across so far at the end of <
http://code.google.com/p/dataincubator/wiki/OpenLibrary> It's not clear to
me how to map them to specific classes - I need help here.

 Ian

We could probably get a good set of journal publisher data from the
CUFTS KB [1] or the old jake data (although it's now quite out of
date).  Not sure of the licensing.

I assume the OL is already loaded with the LC bib records in archive.org?

-Ross.
1. http://cufts2.lib.sfu.ca/knowledgebase/

Hi Ross,

The CUFTS KB looks really useful, plenty of journals and titles in there.

Cheers,

L.

2009/5/6 Ross Singer <rossfsin...@gmail.com>

--
Leigh Dodds
Programme Manager, Talis Platform
Talis
leigh.do...@talis.com
http://www.talis.com

Ross,

Thanks for reminding me of CUFTS... your point about archive.org is well
made.

Another place to watch, (even though the content is also on there too), is
the Biodiversity Heritage Library, which is getting a helping hand from our
friends at OCLC: http://twitter.com/chrisfreeland/statuses/1668746642

Biodiversity Heritage Library is a good source for scientific articles, etc.
if that's your area... mostly historic stuff, but useful for entomologists,
etc...

cheers,

Tom

AFAIK it contains the LC data - it certainly looks like LC data in a lot of
the records (to my untrained eye and through the JSON filter).