SyntaxHighlighter

SyntaxHighlighter

Showing posts with label skos. Show all posts
Showing posts with label skos. Show all posts

Monday, November 15, 2010

IPTC and Semantic Web Technologies - Linked Data, Metadata and Ontology

At the IPTC's most recent face-to-face meeting in Rome, we reviewed our explorations of semantic web technologies for news. The news standards body has been looking at three major areas:


Linked Data
We discussed our work to turn IPTC's subject codes into Linked Data using SKOS concepts and Dublin Core properties. Michael Steidl (Managing Director of the IPTC) was planning to demo the IPTC Linked Data ... but, sadly, Internet access was not working in the hotel! He was, however, able to discuss the proposed collaboration between the IPTC and MINDS on Linked Data for news.

Linking and Mapping

Much of the discussion about IPTC's Linked Data work turned on the difficulties of mapping. In addition to representing the IPTC subject codes in RDF/XML and RDF/Turtle, there was some work done to map from the 17 top level IPTC terms to dbpedia concepts. We quickly figured out that these top level terms are chiefly umbrella terms and so don't map very well to individual dbpedia concepts. The meeting felt that it would be good to map the second level terms, but the problem is that this is quite a lot of work and - as usual - it isn't clear who will do it! We then explored some of the challenges of creating and maintaining the links in Linked Data - that is where a lot of the value, but also much of the investment, lies.

My slides about IPTC's Linked Data work are available on slideshare:

Metadata in HTML - rNews and hNews
Many news providers have created feeds to supply news using IPTC formats such as NITF and NewsML-G2. However, there are an increasing number of consumers of news who only want to work with "pure" web technologies, i.e. HTML rather than XML. So, the IPTC has been looking at the two major paths to represent metadata in HTML - microformats and RDFa.

hNews
I discussed hNews - the microformat for news that was adopted by the community in late 2009 - which builds upon hAtom by adding a few news-specific fields (such as Source and Dateline). As well as explaining how to add microformats to your HTML templates, I provided some statistics that the Associated Press has gathered on adoption. (As of October 2010, we know of about 1,200 sites using hNews, predominantly in North America). See my Prezi on hNews at http://prezi.com/uo5ggdkll8sa/an-introduction-to-hnews/ for more.

rNews
Evan Sandhaus (Semantic Technologist at The New York Times) described rNews - a proposal for an RDFa vocabulary for news. As the names imply, rNews and hNews are similar in intent (news-specific metadata in HTML) but somewhat different in approach. Whereas hNews went through the microformats process, an RDFa vocabulary can be created by anyone. Evan has created an initial rNews draft based somewhat on the NewsML-G2, NITF and hNews models but it is clearly heavily influenced by the needs of the New York Times.

Members of the IPTC's Semantic Web Yahoo! Group can view Evan's rNews draft and are encouraged to discuss it in that email group. At the Rome face-to-face meeting there was quite a lot of interest, but also several issues raised about the details of the first draft. The meeting generally agreed to continue looking at both hNews and rNews, with a view to making a recommendation on both in 2011.

The benefit of getting rNews and hNews adopted by the IPTC is that greater industry support translates into less work for toolmakers: if many news providers support hNews and/or rNews - and do so in very similar ways - then it is easier to build parsers and tools to extract metadata from HTML.

News Ontology
Benoît Sergent of the European Broadcasting Union discussed the work that he and his colleague Jean-Pierre Evain have been doing to create a news ontology, based upon the NewsML-G2 news model. Benoît described how EBU would like to combine the video content that it produces with content from its member organizations and other third parties. If they can represent this information using a flexible, universal model (the news ontology) they could use off-the-shelf tools (such as a triple store) to query, manipulate and recombine that content.

In many ways, this is the most fundamental piece of the semantic web work that the IPTC is undertaking. It is also the least accessible, for many. Members of the IPTC SemWeb group can view a draft of the news ontology and can comment in that email list.

Friday, May 14, 2010

SKOS and Protoge HOWTO

In the IPTC, we are doing some work to figure out how to represent the IPTC Controlled Vocabularies as Linked Data.  We've decided to use SKOS as the RDF Vocabulary.  One of the things we wanted to do was to use a tool that "understands" SKOS.  We decided to look at Protoge for this.  Here are the steps we figured out to make it work (for some values of work):

Download version 4.0.2 from http://protege.stanford.edu/download/registered.html
-          Install it in your PC
-          Add the SKOSed plugin (use Check for plugins... item under File)
-          Add the Pellet Reasoner plugin
-          (you have to restart Protege before the new plugins are active)
-          Add Views to the Individuals tab: SKOSed view -> Inferred Concept Hierarchy + SKOS Usage
-          Then you may load a SKOS vocabulary – but only with narrower and broader relationships, does not work with the ...Transitive variants.
-          Then you should run the Pellet Reasoner against this vocabulary
-          Only then you should see the hierarchy in the Inferred Concept Hierarchy frame.
(JPE later adds "I have declared the narrower and broader properties as "transitive" using protégé and it  works.")

Posting them here, to make it easier for me to find (and maybe to help others).

Friday, May 7, 2010

Recently, I was asked for pointers to introductory material on the Semantic Web, specifically for such topics as N3 and Turtle.  I found "The Semantic Web 1-2-3" the most helpful in starting to get to grips with the mysteries of the SemWeb.  Note, however, that it is somewhat outdated now (it refers to DAML+OIL for example).  But it is still a good foundation and it has more links to great semwebby material than you can shake a stick at, if that's your idea of fun.  A couple of extra links that might be of help are
I can't really find anything good that explains Turtle, other than the formal spec.  But my twitter-length explanation is that Turtle is N3 minus the reasoning extensions but plus internationalization (i.e. it is a more exact rendition of RDF than N3 is).

Other great links out there?

Monday, April 26, 2010

Experimenting with Bull Fighting and the Semantic Web

Paul Kelly and I took a look at what it would take to represent NewsML-G2 style Knowledge Items as SKOS.


 Here's a typical G2 style concept from the IPTC subject code vocabulary:

<concept>
<conceptId qcode="subj:01003000" created="2008-01-01T00:00:00+00:00" />
<type qcode="cpnat:abstract">
<name xml:lang="en-GB">Abstract concept</name>
</type>
<name xml:lang="en-GB">bullfighting</name>
<name xml:lang="de">Stierkampf</name>
<name xml:lang="fr">Tauromachie</name>
<name xml:lang="es">toros</name>
<name xml:lang="es">??</name>
<definition xml:lang="en-GB">Classical contest pitting man against the bull</definition>
<definition xml:lang="de">Klassischer Wettkampf Mann gegen Stier.</definition>
<definition xml:lang="fr">Tauromachie, combat entre un homme et un taureau</definition>
<definition xml:lang="es">Clasico enfrentamiento entre Toro y Hombre.</definition>
<definition xml:lang="es">????????????????</definition>
<broader qcode="subj:01000000" type="cpnat:abstract" />
</concept>

Here's what I think that would render as in SKOS - first in N3:

@prefix dc: <http://purl.org/dc/elements/1.1/>.
@prefix skos: <http://www.w3.org/2004/02/skos/core#>.
@prefix subj: <http://iptcsubj.example.com/>.
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>.

subj:01003000 rdf:type skos:Concept;
dc:created "2008-01-01T00:00:00+00:00";
skos:prefLabel "bullfighting"@en-GB;
skos:prefLabel "Stierkampf"@de;
skos:prefLabel "Tauromachie"@fr;
skos:prefLabel "toros"@es;
skos:prefLabel "闘牛"@es;
skos:definition "Classical contest pitting man against the bull"@en-GB;
skos:definition "Klassischer Wettkampf Mann gegen Stier."@de;
skos:definition "Tauromachie, combat entre un homme et un taureau"@fr;
skos:definition "Clasico enfrentamiento entre Toro y Hombre."@es;
skos:definition "人間を雄牛と戦わせる伝統的な競技"@es;
skos:broader subj:01000000.

Some conclusions. G2 concepts are easy to map to SKOS! But there are some choices that need to be made (e.g. do we use skos:broader or owl:broader?) It feels like a G2 conceptset is almost identical to a SKOS Concept Scheme.

And there's a bug in the IPTC mapping of the subject codes (the Japanese definition and name is labeled as Spanish).


¡Olé! by karwik