The FRBR Blog

Jenn Riley sent MARC work identification algorithm specifications released to the Music Library Association mailing list yesterday:

The Indiana University Variations3 digital music library project has performed a series of experiments designed to maximize the information we can map to our work-based metadata model from MARC Bibliographic and Authority records. The biggest strength of our "batch loading" algorithm is better identification of Works that are represented in bibliographic records, especially in cases where one record represents *multiple* work. The output of our batch loading work is a full mapping from MARC to data that conforms to the current Variations work-based metadata model, although we will be updating our specifications to output fully FRBRized data in the future. The full current algorithm and supporting documents are now available at <http://wiki.dlib.indiana.edu/confluence/x/OQBGBQ>. We hope that even if they aren’t fully FRBRized yet they may be of use to others in the community looking to use MARC records for the basis of FRBRized systems.

The work to migrate these batch loading specifications will be done as part of the newly IMLS-funded Variations/FRBR project at Indiana University. More information on the Variations/FRBR project can be found at <http://www.dlib.indiana.edu/projects/vfrbr/>.

Exciting stuff! I asked Jenn if they had any code to release — the programs they’d written to implement the algorithm — and she said not yet, but they would, along with more details about the XML representation of the data. Work is continuing, and this is definitely something to watch.

(Via Stacy Allison-Cassin.)

Gene Fieg sent some interesting mail to the RDA-L mailing list on Monday. Look through the responses for good comments.

After reading RDA and its application of FRBR, it seems that we dealing with librarianship’s application of Platonism, especially in the descriptions work, expression, manifestation, and item. There really is no “work”; it is like a Platonic form, which is reflected in its physicality by expression, manifestation, and item. We, as catalogers, actually deal with the item. So perhaps in the real world the relationship should be item, manifestation, expression, work. The item points to the manifestation which points to the expression which points to the Platonic ideal, work.

The Stanford Encyclopedia of Philosophy has entries on Platonism in metaphysics and Platonism in mathematics.

Identifying FRBR Work-Level Data in MARC Bibliographic Records for Manifestations of Moving Images, by Kelly McGrath and Lynne Bisko, is in issue 5 of The Code4Lib Journal. Here’s the abstract:

The library metadata community is dealing with the challenge of implementing the conceptual model, Functional Requirements for Bibliographic Records (FRBR). In response, the Online Audiovisual Catalogers (OLAC) created a task force to study the issues related to creating and using FRBR-based work-level records for moving images. This article presents one part of the task force’s work: it looks at the feasibility of creating provisional FRBR work-level records for moving images by extracting data from existing manifestation-level bibliographic records. Using a sample of 941 MARC records, a subgroup of the task force conducted a pilot project to look at five characteristics of moving image works. Here they discuss their methodology; analysis; selected results for two elements, original date (year) and director name; and conclude with some suggested changes to MARC coding and current cataloging policy.

Two from Diane Hillmann with brief mentions of FRBR amidst much about Resource Description and Access:

• Show Me the Data!: “I was reminded of this when hearing a developer (who shall remain nameless) say recently that ‘of course you’d never show anyone the FRBR structure’ in an RDA based system. ‘Why the hell not?’ I responded.”
• Getting There (to be published in the January/February 2009 issue of Technicalities): “MARBI (the MARC standards advisory body) entertained some small proposals to make adjustments to MARC to accommodate RDA, feeding deep denial on the part of those in the cataloging community that fervently hoped that RDA would not require a move to something other than MARC. This was true even as each RDA revision continued to express more explicitly its underlying FRBR model, a model that is not currently accommodated in the MARC record.”

Read the Technicalities article, it’s good.

Resource Description and Access is going to be a web-based tool, and from the demo I saw a few months ago it could be really good. Unfortunately right now rdaonline.org is only giving away enormous PDFs of print versions. This simple HTML version of RDA has the whole thing in a much easier to read format. You can search the rules for words like manifestation. This is a really good, and fast, hack.

A couple of recent posts from Xiaoming Liu, the fellow who looks after xISBN and those other useful services from OCLC:

• Support hathitrust.org in xOCLCNUM Service. “This feature is implemented in following way: a request can put an additional parameter ‘library=hathi’ in xOCLCNUM request, the service will only return records which [are] marked as free access in hathitrust.org.”
• Better Hyphen Support in xISBN Service. “When requested ISBN is hyphenated, the response ISBNs will be hyphenated as well; we also added additional method of explicitly hyphenating ISBNs.”

I wanted lots of Harry Potter ISBNs, so I was doing some superduping. For example, I superduped 1551922460, the ISBN of the 1999 hardcover Raincoast Books manifestation of Harry Potter and the Prisoner of Azkaban.

• xISBN: 103 results
• thingISBN: 141 results

If you combine and dedupe them, you get 169 ISBNs. If you superdupe them, you get 1083. That is, you take the first number from thingISBN that xISBN didn’t tell you about, and ask xISBN about it. That gives you a new set of ISBNs. Do that for all the thingISBN-only numbers, then reverse the process and ask thingISBN about the numbers xISBN told you about. Repeat, back and forth, until you’ve exhausted both sides and pulled all of the ISBNs out of their different partitions and put them into one big bucket.

I did this for all of the Harry Potter books, and after careful examination my keen eyes noticed something:

   ISBNs Title
    1083 isbns-01-philosophers-stone.txt
    1083 isbns-02-chamber-of-secrets.txt
    1083 isbns-03-prisoner-of-azkaban.txt
    1083 isbns-04-goblet-of-fire.txt
    1083 isbns-05-order-of-the-phoenix.txt
    1083 isbns-06-half-blood-prince.txt
     121 isbns-07-deathly-hallows.txt
       3 isbns-0x-beedle.txt
      53 isbns-0x-scamander.txt

Superduping the ISBNs of the first six Harry Potter books had given 1083 ISBNs for each! And sure enough they’re the same 1083 ISBNs. What’s going on here is that because of boxed sets and other collections, and possibly incorrect work-groupings by hand and by algorithm, once you start looking at one Harry Potter book through xISBN and thingISBN, you end up looking at all of them. Or almost all. The seventh one stands alone, but I think that will change in a year or two, and it will fall in with the others.

This work supercluster includes all of the Harry Potter books, the movies, some soundtracks, some scores, some derivative works like pop-up books, and more. It also includes books by Carl Sagan, Philip Pullman, and C.S. Forester (!).

This supercluster phenomenon is interesting. In part it’s caused by collected editions and boxed sets and no easy standard way of handling two works in one manifestation. Human and machine error is also involved. xISBN and thingISBN aren’t perfect, and superduping their results compounds errors from one into the other and you can end up with a bit of a mess.

(I tried superduping Pride and Prejudice and stopped when I started getting into the complete works of Shakespeare. I’ll post more about that if I try it again, but perhaps all the great works of English literature are in one giant confused FRBRy supercluster.)

Full FRBRization, where relationships between works and aggregate works (such as boxed sets and omnibus editions) are clearly specified, will mean this isn’t a problem. That’ll be a lot of work, though.

Using isbn2marc I found MARC records for 978 of the 1234 total ISBNs.

978 Harry Potter-related MARC records (1 MB MARC)

I ran them through the LC FRBRization tool and put them into OpenFRBR.

~/src/openfrbr$ ./script/console
Loading development environment (Rails 2.1.0)
>> Work.find(:all).size
=> 171
>> Expression.find(:all).size
=> 471
>> Manifestation.find(:all).size
=> 973
>> Person.find(:all).size
=> 22
>> Creation.find(:all).size
=> 138

I’ve been hacking away the last few days and got a few good things working in OpenFRBR so I called it 2.1 and put it up.

• Click to edit attributes of works and expressions (but reload the page to see the changes in the heading, because I don’t know how to update two elements at once yet). No Submit button necessary, because I use in_place_editor, some Ajax magic.
• You can reassign an expression to a different work using the select menu at the top of the page. Look at the works list and find a work that has one or two expressions. The name (identifier) will end in “novel/1″ or “novel?/2″ or something like that. Look at the work, then look at its expression, then reassign it. Delete the work with zero expressions, if you want. This makes it easier to fix the mistakes made by an FRBRizing algorithm (like the Library of Congress’s, which I used on 160 or so Harry Potter bib records).
• I updated isbn2marc so it uses a WorldCat API key if you have one. This really speeds things up and makes it easier to get lots of MARC records. I originally wrote isbn2marc because I had no access to WorldCat and needed to poll any open Z39.50 server I could find to get MARC records. Now I do have access to it, which is nice, but the script works as well as ever if you don’t use WorldCat. I think I got MARC records at WorldCat for 140 or 150 ISBNs out of the 180 I had, and I found 10 or 20 more at open Z39.50 servers.

There are character set problems but I”m just ignoring them until the next version of Ruby.

Working on this has given me a better idea of what to do next, but what with Christmas and all I may not get much hacking done the next little while. Some parts of the entity relationships aren’t all fleshed in, but I’ll probably take the Work-Expression relation and expose the relation attribute there and make that all nice and editable. After that I could use the same approach for other entities and relationships, that is, doing for the Embodiment relation (Expression-Manifestation) what I do with Reification (what I call the Work-Expression relation). I haven’t documented the data model yet but you can check the source code, or ask.

OCLC has xISBN, of course. And LibX, the incredibly useful Firefox extension, uses it. If you’re looking at one print manifestation and you know its ISBN, it’s the easy way to get to other manifestations of the same work from your library. (LibX also does much more. Check it out if you don’t use it already.)

Tony Hirst posted Looking Up Alternative Copies of a Book on Amazon, via ThingISBN, wherein he shows, with snazzy graphics, how he uses Yahoo Pipes (which is a really impressive piece of work, and fun to use) and LIbraryThing’s thingISBN to do something similar.

He spoke about it at Mashed Libraries 2008 and Andy McGregor wrote Mashing thingISBN and Library Lookup Using Yahoo Pipes Courtesy of Mashed Libraries 2008