rest, the semantic web and my feeble brain

Imagine you were minting close to a million URIs for historic newspaper pages such as:

for pages like:

The web page allows you to zoom in quite close and see lots of detail in the page:

Now lets say I want to describe this Newspaper Page in RDF. I need to decide what subject URI to hang the description off of. Should I consider this Newspaper Page resource an information resource, or a real world resource? The answer to this question determines whether or not I can hang my description of the page off the above URI, for example:

  dcterms:issued "1898-01-01"^^<> .

Or if I need to mint a new URI for the page as a real world thing:

  dcterms:issued "1898-01-01"^^<> .

AWWW 1 provides some guidance:

By design a URI identifies one resource. We do not limit the scope of what might be a resource. The term “resource” is used in a general sense for whatever might be identified by a URI. It is conventional on the hypertext Web to describe Web pages, images, product catalogs, etc. as “resources”. The distinguishing characteristic of these resources is that all of their essential characteristics can be conveyed in a message. We identify this set as “information resources.”

This document is an example of an information resource. It consists of words and punctuation symbols and graphics and other artifacts that can be encoded, with varying degrees of fidelity, into a sequence of bits. There is nothing about the essential information content of this document that cannot in principle be transfered in a message. In the case of this document, the message payload is the representation of this document.

Can all of the essential characteristics of this newspaper page be sent down the wire as a message to a client? The text of the page is pretty legible after zooming in and you can see pictures, headlines, etc. You can’t feel the texture of the page itself, but you can’t in the microfilm that the page images were generated from. So I’m inclined to say yes.

Cool URIs for the Semantic Web also has some advice:

It is important to understand that using URIs, it is possible to identify both a thing (which may exist outside of the Web) and a Web document describing the thing. For example the person Alice is described on her homepage. Bob may not like the look of the homepage, but fancy the person Alice. So two URIs are needed, one for Alice, one for the homepage or a RDF document describing Alice. The question is where to draw the line between the case where either is possible and the case where only descriptions are available.

According to W3C guidelines ([AWWW], section 2.2.), we have a Web document (there called information resource) if all its essential characteristics can be conveyed in a message. Examples are a Web page, an image or a product catalog.

In HTTP, because a 200 response code should be sent when a Web document has been accessed, but a different setup is needed when publishing URIs that are meant to identify entities which are not Web documents.

This makes me think that I will need distinct identifiers for the abstract notion of the Newspaper Page, and the HTML document itself, if it is important to describe them separately. Say for example if I wanted to say the publisher of the web page was the Library of Congress, but the publisher of the Newspaper Page was Charles M. Shortridge. If I don’t have distinct identifiers I will have to say:

  dc:publisher <>, 

Pondering this Information Resource Sniff-Test got me re-reading Xiaoshu Wang’s paper URI Identity and Web Architecture Revisited again. And I’ve come away more convinced that maybe he’s right: that the real issue lies in my vocabulary usage (dc:publisher in this example), and not with whether my URI identifies an Information Resource or not. So maybe new vocabulary is needed in order to describe the representation?

  web:repPublisher <> ;
  dcterms:publisher <> 

But there isn’t a community of practice behind Xiaoshu’s position, at least not one like the Linked Data community. Unless perhaps his position is closer to the REST community which is going strong at the moment, especially in AtomPub circles. Members of the linked-data/semweb community would most likely say that there needs to be either hash or 303’ing URIs for the Newspaper Page, distinct from the URIs for the document describing the Newspaper Page. As a late comer to the httpRange-14 debate I don’t think I ever internalized how REST and the Semantic Web are slightly out of tune w/ each other regarding resources on the web.

So. Should I have two different URIs: one for the real-world Newspaper Page, and one for the HTML document that describes that page? Is the Newspaper Page an Information Resource? Am I muddling up something here? Am I thinking too much? Should I just let sleeping dogs lie? Your opinion, advice, therapy would be greatly appreciated.

VocabularySoup (1)

It’s been great to see RDFa being picked up by web2.0 publishers like Digg and MySpace. You can use the RDFa Distiller to extract the RDFa from a given web page u by constructing a URI like:

Which translates kind of nicely into a command line utility to add to your ~/bin:

curl "$1"

So with that little shell script in hand I can now look at the RDFa something like Yo La Tengo’s page on MySpace:

ed@rorty:~$ rdfa

LibraryThing Ubuntu Screen Saver

I read about the LibraryThing Mac Screensaver and of course wanted the same thing for my Ubuntu workstation at $work. Naturally, I’m supposed to be working on some high-priority tickets on a tight deadline…so I started to work right away on how to do this. Your tax dollars at work, etc…

I’m sure that there’s a much more elegant way of doing this, but I basically created a simple python program extract-images that will pull image urls out of arbitrary text, suck down the images, and dump them to a directory. This can be combined with cron and the standard GLSlideshow screensaver, which displays a slideshow of images in a particular directory.

So you just download extract-images, put it in your path, add a crontab entry like (substituting edsu for your LibraryThing username):

00 14 * * * extract-images /home/ed/Pictures/covers

And then tell GLSlideshow where your images are by adding this to your ~/.xscreensaver

imageDirectory:   /home/ed/Pictures/covers
chooseRandomImages:   True

Dear $manager, it really didn’t take me that long to do this. Honest!

APIs Suck

With TransparencyCamp last weekend, news of the mandated use of feed syndication by Federal Agencies receiving funds from the Recovery Act, recent blog posts by Tim O’Reilly and the Special Libraries Association, an article in Newsweek, news of Carl Malamud’s bid to become the Public Printer of the United States (aka head of the GPO), and the W3C eGov meeting coming up next week it looks like issues related public access to government data (specifically Library of Congress bibliographic and legislative data) are hitting the mainstream media, and getting political mind-share. Exciting times.

One thing that bubbled up at code4lib2009 last week was the notion that APIs Suck. Not that web2.0 APIs are wrong or bad…they’re actually great, especially when compared to a world where no machine access to the data existed before. The point is that sometimes just having access to the raw data in the ‘lowest level format’ is the ideal. Rather than service providers trying to guess what you are trying to do with their data, and absorbing the computational responsibility of delivering it, why not make the data readily available using a protocol like HTTP? Put the data in a directory, turn on Indexes, do some sensible caching, and maybe gzip compression and let people grab it, and robots crawl it. Or maybe use something like Amazon Public Datasets. It seems like a relatively easy first step, that involves very little custom software development, and one with the ability make a huge impact.

I’m a federal employee, so I really can’t come out and formally advocate directly for political appointments. But I have to say it would great to see someone like Malamud at the helm of the GPO, since he’s been doing just this kind of work for 20 years. Exciting times.


So code4lib2009 was a whole lot of fun. The amazing thing about the conference isn’t really reflected in the program of talks. I feel like I can say that since I was one of them.

The real value is the social space and the time to talk to people you’ve seen online, throw around ideas, get background/contextual information on projects, etc. Hats off to Jean Rainwater and Birkin Diana for picking an beautifully casual and intimate hotel to hold the conference in.

It’s taken me a few days to get some perspective on all that happened. In the meantime I’ve read a few accounts that capture important aspects of the event from: Terry Reese, Jon Phipps, Jay Luker, Declan Fleming, Richard Wallis (1,2,3), Dan Chudnov, Gabe Farrell.

The Linked Data Pre-conference was quite valuable. For one it gave attendees some experience in what it means to publish data in a distributed way, and to write code to aggregate it using a attendees/FOAF experiment. Mike Giarlo aptly surmised from this that the key points for teaching beginners about linked data are that:

  1. Validators are essential
  2. You are not your FOAF

In other words:

  1. Am I doing this rdf/xml, turtle, rdfa right?
  2. ZOMG, httpRange-14!

Ian Davis presented the basics of RDF for people who are already familiar with traditional data management. Apparently Ian’s slides hit #1 for the day on SlideShare, which highlights the interest in linked data that is percolating through the Web. The pre-conf was very well attended as well.

Some folks like Jonathan Brinley and Michael Klein were able to hack on a Supybot Plugin to work with the FOAF data generated by the crawler. I also got chatting with William Denton about the potential of linked data for FRBR/RDA efforts. Unfortunately I didn’t hear about Alistair Miles’ new project on google-code for exploring the translation of traditional MARC/MODS into RDA/FRBR until after the event. Most of the other slides from presenters at the pre-conf are available from the wiki page.

I was really struck by some of the issues that Dan Chudnov raised in his talk about Caching and Proxying Linked Data right before lunch. In particular his comparison of the Linking Open Data Cloud to what libraries understand as their ready reference collection:

See p.9 of Dan’s slides

Dan explored how we need to think about the technical and administrative details of managing linked-data if linked-data is to be taken seriously by the library community. Relatedly the pre-conf gave me an opportunity to publicly apologize to Anders Söderbäck for yanking offline in such an abrupt manner, and disturbing his links from subject authority records at to Dan’s ideas for consuming library linked data and Anders and mine experience publishing library linked data gelled nicely in my brain. Similar ideas from Jon Phipps (one of the authors of Best Practice Recipes for Publishing RDF Vocabularies) have led me to believe this could be a nice little area for some research.

Prepping for the pre-conference itself was good fun, since it led me to discover a series of connections between the early development of the www and Brown University (where the conference was being held) and the history of hyperdata/text: in a nutshell it was Tim Berners-Lee’s proposal for the web -> Dynatext -> Steve DeRose -> Andy van Dam -> Hypertext Editing System -> Ted Nelson -> Doug Engelbart -> Vannevar Bush. Yeah, I guess you had to be there … or maybe that didn’t help. At any rate the slides, complete with breakdancing instructions are available.

I haven’t even started talking about the main event yet. The things I took away from the 3 days of presentations and talks, in no particular order were:

  • I want to learn more about the Author-ID effort that Geoffrey Bilder talked about
  • Stefano Mazzocchi’s keynote and Sean Hannan’s presentation convinced me that I need to understand and play with Freebase’s JavaScript application development environment Acre and the sparql-ish, query by example Metaweb Query Language (MQL). It seems like Freebase is exploring some really interesting territory in building a shared knowledge base of machine readable, human editable data, which can sit behind a seemingly infinite amount of web presentation layers.
  • Terence Ingram’s presentation, Ross Singer’s presentation about Jangle, me and Mike’s SWORD presentation, and a chat with Fedora/REST proponent Matt Zumwalt, and hearing about the Talis Platform have convinced me that real REST has got mind-share and traction in the library technology world.
  • Ian Davis’ keynote on the second day captured for me, the constant challenge it is to stay true to the roots of the web, and how important it is to stay true to them. It was really interesting to hear how he emphasized the importance of data over code, and the necessity for decentralization compared with the centralization.
  • Chatting with Jodi Schneider and William Denton and listening to their presentation made me want to understand RDA and FRBR at a practical level. This includes getting into the vocabularies that are being developed, and trying to convert some data. The history of FRBR in particular as told by Bill is also a gateway into a really fascinating history of cataloging. Also the work that Diane Hillman and Jon Phipps have been doing to enable vocabulary development like RDA/FRBR seems really important to keep abreast of.

More tidbits will probably float into my blog or into my tweets over the coming weeks, as the beer wears off, and the ideas sink in. But for now I’ll leave you with some of my favorite photos from the conference. It’s the people that makes code4lib what it is. It was great to connect up, and meet new folks in the field.


Oh and in case you missed it, the tweetstream and the other fine photos.

the importance of being crawled

While was up and running harvesters actively crawled it. At its core all did was mint a URI for every Library of Congress Subject Heading. This is similar in spirit to Brewster Kahle’s more ambitious OpenLibrary project to mint a URI for every book, or in his words:

One web page for every book

Aside: It’s also similar in spirit to RESTful web development, and to the linked data, semantic web effort generally.

Minting a URI for every Library of Congress Subject Heading meant that there were lots of densely interlinked pages. Some researchers at Stanford did a data visualization of LCSH two years ago, which illustrates just how deeply linked LCSH is:

I wanted to get crawled so I intentionally put some high level, well connected concepts (Humanities, Science, etc) on the home page to provide a doorway for web crawlers to walk through into the site and begin discovering all the broader, narrower, related links between concepts–without having to perform a search.

So is down now, but it turns out you can still see its shadow living on in quite a usable form in web search engines. For example type this into any of the big three search engines: mathematics

And you’ll see:




It’s interesting that (unlike Google and Yahoo) Microsoft’s relevancy ranking actually puts the heading for “Mathematics” at the top. Also note that simple things like giving the page a good title, and descriptive text make the heading show up in usable form in each search engine.

It’s not too surprising that trying the same for doesn’t work out so well. Umm, yeah

On the one hand, I’m just being nostalgic looking at the content that once was there &sigh;. But on the other there seems to be a powerful message here, that putting data out onto the open web, and making it crawlable means your content is viewable via lots of different lenses. Maybe you don’t have to get search exactly right on your website, let other people do it for you.

Two other things come to mind: LOCKSS and Brewster’s even more ambitious project. I’ve been sort hoping that somehow or another the Internet Archive and the Open Library would find there way into being publicly funded projects. What if? I can daydream right?

crawling bibliographic data

Today’s Guardian article Why you can’t find a library book in your search engine prompted me to look at Worldcat’s robots.txt file for the first time. Part of the beauty of the web is that it’s an open information space where anyone (people and robots) can start with a single URL and follow their nose to other URLs. This seemingly simple principle is what has allowed a advertising^w search company like Google (that we all use every day) to grow and prosper.

The robots.txt file is a simple mechanism that allows web publishers to tell web crawlers what they are allowed to look at on a website. Predictably, the files are always found at the root of a website in a file named robots.txt. You don’t have to have one, but many publishers like to control what gets indexed on their website, sometimes to hide content, and other times to shield what may be costly server side operations. Anyway, here’s what you see today for

User-agent: *
Disallow: /search


So this instructs a web crawler to not follow any links that match /search in the path, such as:

Now if you look on the homepage for Worldcat there are very few links into the dense bibliographic information space that is worldcat. But you’ll notice a few in the lower left box “Create lists”. So a crawler could for example discover a link to:

This URL is allowed by the robots.txt so the harvester could go on to that page. Once at that item page there are lots of links to other bibliographic records: but notice the ones to other record displays all seem to match the /search pattern disallowed by the robots.txt, such as:


So a web crawler will not be able to wander into the rich syndetic structure of Worldcat and start indexing.

However, all is not lost. Notice above that OCLC does reference a Worldcat sitemap in their robots.txt. Sitemaps are a lightweight mechanism that Yahoo, Google and Microsoft developed for instructing a web harvester on how to walk through a site.

So if we look at OCLC’s sitemap sitemap we’ll see this:

< ?xml version="1.0" encoding="UTF-8"?>


This essentially defers to two other sitemaps. The first 30 lines of the first one (careful in clicking it’s big!) looks like:

< ?xml version="1.0" encoding="UTF-8"?>




Now we can see the beauty of sitemaps. They are basically just an XML representation for sets of web resources, much like syndicated feeds. There are actually 40,000 links listed in the first sitemap file, and 12,496 in the second. Now URLs like

are clearly allowed by the robots.txt file. So indexers can wander around and index the lovely identities portion of Worldcat. It’s interesting though, that the content served up by the identities portion of Worldcat is not HTML–it’s XML that’s transformed client side to HTML w/ XSLT. So it’s unclear how much a stock web crawler would be able to discover from the XML. If google/yahoo/microsoft’s crawlers are able to apply the XSLT transform, they will get some HTML to chew on. But notice in the HTML view that all the links into Worldcat proper (that aren’t other identities) are disallowed because they start with /search.

And a quick grep and perl pipeline confirm that all 52496 urls in the sitemap are to the identies portion of the site…

So this is a long way of asking: I wonder if web crawlers are crawling the books views on Worldcat at all? I imagine someone else has written about this already, and there is a known answer, but I felt like writing about the web and library data anyhow.

Since OCLC has gone through the effort of providing a web presentation for millions of books, and even links out to the libraries that hold them, they seem uniquely positioned to provide a global gateway for web crawlers to the library catalogs around the world. The links from worldcat out to the rest of the world’s catalogs would turn OCLC into a bibliographic super node in the graph of the web, much like Amazon and Google Books. But perhaps this is perceived as giving up the family jewels? Or maybe it would put to much stress on the system? Of course it would also be great to see machine readable data served up in a similar linked way

So in conclusion, it to would be awesome to see either (or maybe both):

  • the /search exclusion removed from the robots.txt file
  • sitemaps added for the web resources that look like

Of course one of the big projects I work on at LC is Chronicling America which is currently excluded by LC’s robots.txt…so I know that there can be real reasons for restricting crawling access (in our case performance problems we are trying to fix).

Oh gosh, I just noticed when re-reading the Guardian article that my experiment was mentioned. Hopefully there will be good news to report from LC on this front shortly.

work identifiers and the web

Michael Smethurst’s In Search of Cultural Identifiers post over at the BBC Radio Labs got me thinking about web identifiers for works, about LibraryThing and OCLC as linked library data providers, and finally about the International Standard Text Code. Admittedly it’s kind of a hodge-podge of topics, and I’m going to taking some liberties with what ‘linked data’ and ‘works’ mean, so bear with me.

Both OCLC Worldcat and LibraryThing mint URIs for bibliographic works, like these for Wide Sargasso Sea:

So the library community really does have web identifiers for works–or more precisely web identifiers for human readable records about works. What’s missing (IMHO) is the ability to use that identifier to get back something meaningful for a machine. Tools like Zotero need to scrape the screen to pull out the data points of interest to citation management. Sure, if you want you can implement COinS or unAPI to allow the metadata to be extracted, but could there be a more web-friendly way of doing this?

Consider how blog syndication works on the web. You visit a blog (like this one) and your browser is able to magically figure out the location of an RSS or Atom feed for the blog, and give you an option to subscribe to it.

Well it’s not really magic it’s just a bit of markup in the HTML:

Simple right?

Now back to work identifiers. Consider that both Worldcat and LibraryThing have web2.0 apis for retrieving machine readable data for a work:{work_id}&apikey={your_key}


What if the web pages for these resources at OCLC and LibraryThing linked directly to these machine readable versions? For example if the page for Wide Sargasso Sea at LibraryThing contained this in its <head> element:

This would allow browsers, plugin tools like Zotero and web crawlers to follow the natural grain of the web and discover the machine readable representation. Admittedly this is something that COinS and unapi are designed to do. But the COinS and unAPI protocols are really optimized for making citation data, and non web identifiers available and routable via a resolver of some kind. Maybe I’m just over reaching a bit, but this approach of using the <link> header seems to embrace the notion that there are resources within the Worldcat and Librarything websites, and there can be alternate representations of those resources that can be discovered in a hypertext-driven way.

Of course there is the issue of the API key. In the example above I used the demo key in LibraryThing’s docs. More important in the context of web identifiers for works is the need to distinguish between the identifier for the record, and the identifier for the concept of the work, which is most elegantly solved (IMHO) by following a pattern from the Cool URIs for the Semantic Web doc. But I think it’s important that people realize that it’s not necessary to jump headlong into RDF to start leveraging some of the principles behind the Architecture of the World Wide Web. Henry Thompson has a nice web-centric discussion of this issue in his What’s a URI and Why Does it Matter?

While writing this blog post I noticed a thread over on Autocat that Bowker has been named the US Registrar for the International Standard Text Code. The gist is that the ISTC will be a “global identification system for textual works”, and that registrars (like Bowker) will mint identifiers for works, such as:

ISTC 0A9 2002 12B4A105 7

Where the structure of the identifier is roughly:

ISTC {registration agency} {year element} {work element} {check digit}

It’s interesting that the meat of the ISTC is the work element that is:

… assigned automatically by the central ISTC registration system after a metadata record has been submitted for registration and the system has verified that the record is unique;

The metadata record in question is actually a chunk of ONIX, which presumably Bowker will send to the ISTC central registrar, and get back a work id.

This work that the ISTC is taking on is really important–and one would imagine quite costly. One thing I would suggest to them is that they may want to make the ISTC codes have a URI equivalent like:


They also should encourage Bowker and other registrars to publish their work identifiers on the web:

It seems to me that we might (in the long term) be better served by a system that embraces the distributed nature of the web. A web in which organizations like Bowker, ISTC, OCLC, LibraryThing, Library of Congress and national libraries publish their work identifiers using URIs, and return meaningful metadata for them. Rather than waiting for other people to solve our problems, why don’t we start solving them ourselves bottom-up instead of waiting for someone else to solve it top-down?

Anyhow I feel like I’m kind of being messy in suggesting this linked-data-lite idea. Is it heresy? My alibi/excuse is that I’ve been sitting in the same room as dchud for extended periods of time.

q & a

Q: What do 100 year old knitting patterns and a lost Robert Louis-Stevenson story have in common?

A: A digitally preserved newspaper page.

Q: What about if you add:

A: Just a typical lunch time conversation at Pete’s with a couple people I work with. The cool thing (for me) is that this is normal, involves a host of smart/interesting characters, and is routinely encouraged. I love my job.

100,000 Books and FRBR

The news about 100,000 books on Freebase got me poking around with curl. I was pleased to see that Freebase actually distinguishes between a book as a work, and a particular edition of that book. To FRBR aficionados this will be familiar as the difference between a Work and a Manifestation:

For example here is a URI for James Joyce’s Dubliners as a work:

and here is a URI for a 1991 edition of Dubliners:

If you follow those links in your browser you’ll most likely be redirected to the human readable html view. But machine agents can use the same URL to discover say an RDF representation of this edition of Dubliners, for example with curl:

curl --location --header "Accept: application/turtle"

@prefix fb:
@prefix rdf:
@prefix rdfs:
@prefix xml:

 <> a 
     <> "0486268705";
     <> "91008517";
     <> <>;
     <> <>;
     <> "823";
     <> <>;
     <> "1991";
     <> "Dubliners";
     <> <>. 

 <> a <>;
     <> "152"^^<>;
     <> <>. 

There are a few assertions that struck me as interesting:

  • the statement in red that states that the resource is in fact an edition (of type
  • the statement in green which links the edition with the work (
  • and the assertion in blue which states the Library of Congress Control Number (LCCN) for the book

I was mostly surprised to see the library-centric metadata being collected such as LCCN, OCLC Number, Dewey Decimal Classification, LC Classification. There are even human readable instructions for how to enter the data (take that AACR2!).

Anyhow it got me wondering what it would be like to stuff all the Freebase book data into a triple store, assert:

<> <owl:sameAs> <> .
<> <owl:sameAs> <> .

and then run some basic inferencing and get some FRBR data. I know, crazy-talk … but it’s interesting in theory (to me at least).