larlet-fr-david-cache/cache/8df6b7af9ac944275a13f0d0e97ad7d7/index.md

title: Linked Data - Design Issues
url: https://www.w3.org/DesignIssues/LinkedData.html
hash_url: 8df6b7af9ac944275a13f0d0e97ad7d7


<a href="http://www.cafepress.com/w3c_shop"><img alt="Get a 5* mug" border="none" src="https://www.w3.org/DesignIssues/diagrams/lod/597992118v2_350x350_Back.jpg" align="right"/></a>

<h1>Linked Data</h1>

<p>The Semantic Web isn't just about putting data on the web. It
is about making links, so that a person or machine can explore
the web of data.  With linked data, when you have some of
it, you can find other, related, data.</p>

<p>Like the web of hypertext, the web of data is constructed with
documents on the web. However,  unlike the web of hypertext,
 where links are relationships anchors in
hypertext documents written in <small>HTML</small>, for data they
links  between arbitrary things described by
<small>RDF</small>,.  The <small>URI</small>s identify any
kind of object or  concept.   But for
<small>HTML</small> or <small>RDF</small>, the same expectations
apply to make the web grow:</p>

<ol>
  <li>
    <p>Use <small>URI</small>s as names for things</p>
  </li>

  <li>
    <p>Use <small>HTTP</small> <small>URI</small>s so that people
    can look up those names.</p>
  </li>

  <li>
    <p>When someone looks up a <small>URI</small>, provide useful
    information, using the standards (RDF*, SPARQL)</p>
  </li>

  <li>
    <p>Include links to other <small>URIs</small>. so that they
    can discover more things.</p>
  </li>
</ol>

<p>Simple.  In fact, though, a surprising amount of data
isn't linked in 2006, because of problems with one or more of the
steps.  This article discusses solutions to these problems,
details of implementation, and factors affecting choices about
how you publish your data.</p>

<h2>The four rules</h2>

<p>I'll refer to the steps above as rules, but they are
expectations of behavior.  Breaking them does not destroy
anything, but misses an opportunity to make  data
interconnected.  This in turn limits the ways it can later
be reused in unexpected ways.  It is the unexpected re-use
of information which is the value added by the web.</p>

<p>The first rule, to identify things with
<small>URI</small>s,  is pretty much understood by most
people doing semantic web technology.  If it doesn't use the
universal <small>URI</small> set of symbols, we don't call it
Semantic Web.<br/>
<br/>
The second rule, to use <small>HTTP</small>
<small>URI</small>s,  is also widely understood.  The
only deviation has been, since the web started,  a constant
tendency for people to invent new <small>URI</small> schemes (and
sub-schemes within the <span>urn:</span> scheme)  such as
<small>LSID</small>s and handles and <small>XRI</small>s and
<small>DOI</small>s and so on, for various reasons.
 Typically, these involve not wanting to commit to the
established Domain Name System (<small>DNS</small>) for
delegation of authority but to construct something under separate
control.   Sometimes it has to do with not understanding
that <small>HTTP</small> <small>URI</small>s are names (not
addresses) and that <small>HTTP</small> name lookup is a complex,
powerful and evolving set of standards. This issue discussed at
length elsewhere, and time does not allow us to delve into it
here. [ @@ref TAG finding, etc])</p>

<p>The third rule, that one should serve information on the web
against a <small>URI</small>, is, in 2006, well followed for most
ontologies, but, for some reason, not for some major datasets.
 One can,  in general,  look up the properties and
classes one finds in data, and get information from the
<small>RDF</small>, <small>RDFS</small>, and <small>OWL</small>
ontologies including the relationships between the terms in the
ontology.</p>

<p>The basic format here for RDF/XML, with its popular
alternative serialization N3 (or Turtle). Large datasets provide
a SPARQL query service, but the basic linked data should br
provided as well.</p>

<p>Many research and evaluation projects in the few years of the
Semantic Web technologies produced ontologies, and significant
data stores, but the data, if available at all, is buried in a
zip archive somewhere, rather than being accessible on the web as
linked data.  The Biopax project, the CSAktive data on
computer science research people and projects were two examples.
[The CSAktive data is now (2007) available as linked data]</p>

<p>There is also a large and increasing amount of
<small>URI</small>s of non-ontology data which can be looked up.
 <a href="http://ontoworld.org/wiki/Semantic_wiki">Semantic
wikis</a> are one example. The "Friend of a friend"
(<small>FOAF</small>) and <span>Description of a Project</span>
(<small>DOAP</small>) ontologies are used to build social
networks across the web.    Typical <a href="http://en.wikipedia.org/wiki/List_of_social_networking_websites">
social network portals</a> do not provide links to other sites,
nor expose their data in a standard form.</p>

<p>LiveJournal and Opera Community are two portal web sites which
do in fact publish their data in <small>RDF</small> on the web.
  (Plaxo has a trail scheme, and I'm not sure
whether they support <span>knows</span> links). This means that I can
write in my <small>FOAF</small> file that I know Håkon Lie by
using his <small>URI</small> in the Opera Community data, and a
person or machine browsing that data can then follow that link
and find all his friends. <i>[Update:]</i> Also, the Opera
Community site allows you to register the RDF URI for yourelf on
another site. This means that public data about you from
different sites can be linked together into one web, and a person
or machine starting with your Opera identity can find the others.
</p>

<p>The fourth rule, to make links elsewhere,  is necessary
to connect the data we have into a web, a serious, unbounded web
in which one can find al kinds of things,  just as on the
hypertext web we have managed to build.</p>

<p>In hypertext web sites it is considered generally rather bad
etiquette not to link to related external material.  The
value of your own information is very much a function of what it
links to, as well as the inherent value of the information within
the web page.  So it is also in the Semantic Web.</p>

<p>So let's look at the ways of linking data, starting with the
simplest way of making a link.</p>

<h3>Basic web look-up</h3>

<p>The simplest way to make linked data is to use, in one file, a
<small>URI</small> which points into another.</p>

<p>When you write an <small>RDF</small> file,   say
&lt;http://example.org/smith&gt;, then you can use local
identifiers within the file, say  #albert, #brian and
#carol.  In N3 you might say</p>
<pre>
&lt;#albert&gt;  fam:child &lt;#brian&gt;, &lt;#carol&gt;.
</pre>

<p>or in <small>RDF/XML</small></p>
<pre>
&lt;rdf:Description about="#albert"<br/> &lt;fam:child rdf:Resource="#brian"&gt;<br/>  &lt;fam:child rdf:Resource="#carol"&gt;<br/>&lt;/rdf:Description&gt;
</pre>

<p>The <small>WWW</small> architecture now gives a global
identifier  "http://example.org/smith#albert" to Albert.
 This is a valuable thing to do, as anyone on the planet can
now use that global identifier to refer to Albert and give more
information. </p>

<p>For example, in the
document &lt;http://example.org/jones&gt; someone might
write:</p>
<pre>
&lt;#denise&gt;  fam:child &lt;#edwin&gt;, &lt;smith#carol&gt;.
</pre>

<p>or in <small>RDF/XML</small></p>
<pre>
&lt;rdf:Description about="#denise"<br/> &lt;fam:child rdf:Resource="#edwin"&gt;<br/>  &lt;fam:child rdf:Resource="http://example.org/smith#carol"&gt;<br/>&lt;/rdf:Description&gt;
</pre>

<p><br/>
Clearly it is reasonable for anyone who comes across the
identifier 'http://example.org/smith#carol" to:</p>

<ol>
  <li>Form the <small>URI</small> of the document by truncating
  before the hash</li>

  <li>Access the document to obtain information about #carol</li>
</ol>

<p>We call this dereferencing the <small>URI</small>.  This
is basic semantic web. </p>

<p>There are several variations.</p>

<h3>Variation: URIs without Slashes and HTTP 303</h3>

<p>There are some circumstances in which dividing identifiers
into documents doesn't work very well.   There may logically
be one global symbol per document per document, and there is a
reluctance to include a # in the <small>URI</small> such
as </p>

<p>
http://wordnet.example.net/antidisesablishmentarianism#word</p>Historically,
the early Dublin Core and <small>FOAF</small> vocabularies did
not have # in their URIs.   In any event when
<small>HTTP</small> <small>URI</small>s without hashes are used
for abstract concepts, and there is a document that carries
information about them, then:<br/>

<ol>
  <li>An <small>HTTP</small> <small>GET</small>  request on
  the <small>URI</small> of the concept returns <span>303 See Also</span> and gives in the
  Location: header, the <small>URI</small> of the
  document.  </li>

  <li>The document is retrieved as normal</li>
</ol>

<p>This method has the advantage that <small>URI</small>s can be
made up of all forms.  It has the disadvantage that an
<small>HTTP</small> request mBrowse-ableust be made for every
single one.  In the case of Dublin Core, for example,
dc:title and dc:creator etc are in fact served by the same
ontology document, but  one does not know until they have
each been fetched and returned HTTP redirections.</p>

<h3>Variation: FOAF and rdfs:seeAlso</h3>

<p>The <a href="http://foaf-project.org/">Friend-Of-A-Friend</a> convention
uses a form of data link, but  not using either of the two
forms mentioned above.  To refer to another person in a
<small>FOAF</small> file, the convention was to give two
properties, one pointing to the document they are described in,
and the other for identifying them within that document.</p>
<pre>
&lt;#i&gt;  foaf:knows  [<br/>       foaf:mbox &lt;mailto:joe@example.com&gt;;<br/>        rdfs:seeAlso &lt;http://example.com/foaf/joe&gt; ].
</pre>

<p>Read, "I know that which has email  joe@example.com and
about which more information is in
&lt;http://example.com/foafjoe&gt;".</p>

<p>In fact, for privacy, often people don't put their email
addresses on the web directly, but in fact put a one-way hash
(<small>SHA-1</small>) of their email address and give that. This
clever trick allows people who know their email address already
to work out that it is the same person, without giving the email
away to others.</p>
<pre>
&lt;#i&gt;  foaf:knows  [<br/>       foaf:mbox_sha1sum "2738167846123764823647";  # @@ dummy<br/>  rdfs:seeAslo &lt;http://example.com/foaf/joe&gt; ].
</pre>

<p>This linking system was very successful, forming a
 growing social network, and dominating, in 2006, the linked
data available on the web.</p>

<p>However, the system has the snag that it does not give
<small>URI</small>s to people, and so basic links to them cannot
be made.</p>

<p>I  recommend (e.g in weblogs on <a href="http://dig.csail.mit.edu/breadcrumbs/node/62">Links on the
Semantic Web</a> , <a href="http://dig.csail.mit.edu/breadcrumbs/node/71">Give yourself a
URI</a>, and and <a href="http://dig.csail.mit.edu/breadcrumbs/node/72">Backward and
Forward links in RDF just as important</a>) that those making a
<small>FOAF</small> file give themselves a <small>URI</small> as
well as using the <small>FOAF</small> convention.   
 Similarly, when you refer to a <small>FOAF</small>
 file which gives  a <small>URI</small> to a person,
use it in your reference to that person, so that clients which
just use <small>URI</small>s and don't know about the
<small>FOAF</small> convention can follow the link.</p>

So now we have looked at ways of making a link,
let's look at the  choices of when to make a link.<br/>

<p>One important pattern is a set of data which you can explore
as you go link by link by fetching data.   Whenever one
looks up the URI for a node in the RDF graph, the server returns
information about the arcs out of that node, and the arcs in.
 In other words, it returns any RDF statements in which the
term appears as either subject or object.</p>

<p>Formally,  call a graph G <span>browsable</span> if, for  the URI of
any node in G, if I look up that URI I will be returned
information which describes the node, where describing a node
means:</p>

<ol>
  <li>Returning all statements where the node is a subject or
  object; and</li>

  <li>Describing all blank nodes attached to the node by one
  arc.</li>
</ol><br/>

<p class="detail">(The subgraph returned has been referred to as
"minimum Spanning Graph (MSG [@@ref] ) or  RDF molecule
[@@ref], depending on whether nodes are considered identified if
they can be expressed as a path of function, or reverse inverse
functional properties. A concise bounded description, which only
follows links from subject to object,  does not work.)</p>

<p>In practice, when data is stored in two documents, this means
that any <small>RDF</small> statements which relate things in the
two files must be repeated in each.  So, for example, in my
<small>FOAF</small> page I mention that I am a member of the
<small>DIG</small> group, and that information is repeated on the
<small>DIG</small> group data. Thus, someone starting from the
concept of the group can also find out that I am a member.
 In fact, someone who starts off with my <small>URI</small>
can find all the people who are in the same group.</p>

<h3>Limitations on browseable data</h3>

<p>So statements which relate things in the two documents must be
repeated in each. This clearly is against the first rule of data
storage: don't store the same data in two different places: you
will have problems keeping it consistent.  This is indeed an
issue with browsable data.   A set of  of completely
browsable data with links in both directions has to be completely
consistent, and that takes coordination, especially if different
authors or different programs are involved.</p>

<p>We can have completely browsable data, however, where it is
automatically generated.  The <a href="http://dig.csail.mit.edu/2006/dbview/dbview.py">dbview</a>
 server, for example,  provides a browsable virtual
 documents containing the data from any arbitrary relational
database.</p>

<p>When we have a data from multiple sources, then we have
compromises.  These are often settled by common sense,
asking the question,</p>

<blockquote>
  <p>"If someone has the URI of that thing, what relationships to
  what other objects is it useful to know about?"</p>
</blockquote>

<p>Sometimes, social questions  determine the answer.
 I have links in my <small>FOAF</small> file that I know
various people.  They don't generally repeat that
information in their <small>FOAF</small> files. Someone may say
that they know me, which is an assertion which, in the
<small>FOAF</small> convention, is theirs to assert, and the
reader's to trust or not.  </p>

<p>Other times, the number of arcs makes it impractical.   A
<small>GPS</small> track gives thousands of times at which my
latitude, longitude are known. Every person loading my
<small>FOAF</small> file can expect to get my business card
information, but not all those trackpoints. It is reasonable to
have a pointer from the track (or even each point) to the person
whose position is represented, but not the other way. </p>

<p>One pattern is to have links of a certain property in a
separate document.   A person's homepage doesn't list all
their publications, but instead puts a link to it a separate
document listing them.  There is an understanding
that <span>foaf:made</span>
gives a work of some sort, but <span>foaf:pubs</span> points to a document
giving a list of works.  Thus, someone searching for
something <span>foaf:made</span>
link would do well to follow a <span>foaf:pubs</span> link.  It might
be useful to formalize the notion with a statement like</p>
<pre>
foaf:made  link:listDocumentProperty foaf:pubs.
</pre>

<p>in one of the ontologies.</p>

<h3>Query services</h3>

<p>Sometimes the sheer volume of data makes serving it as lots of
files possible, but cumbersome for efficient remote queries over
the dataset.  In this case, it seems reasonable to provide a
<small>SPARQL</small> query service.  To make the data be
effectively linked, someone who only has the
 <small>URI</small> of something must be able to find their
way the <small>SPARQL</small> endpoint. </p>

<p>Here again the <small>HTTP</small> 303 response can be used,
to refer the enquirer to a document with metadata about which
query service endpoints can provide what information about which
 classes of <small>URI</small>s.</p>Vocabularies for doing
this have not yet been standardized.<br/>

(Added 2010). This year, in order to encourage
people -- especially government data owners -- along the road to
good linked data, I have developped this star rating system.

<p>Linked Data is defined above. Linked <em>Open</em> Data (LOD)
is Linked Data which is released under an open licence, which
does not impede its reuse for free. Creative Commons CC-BY is an
example open licence, as is the UK's <a href="http://www.nationalarchives.gov.uk/doc/open-government-licence/">
Open Government Licence</a>. Linked Data does not of course in
general have to be open -- there is a lot of important use of
lnked data internally, and for personal and group-wide data. You
can have 5-star Linked Data without it being open. However, if it
claims to be Linked Open Data then it does have to be open, to
get any star at all.</p>Under the star scheme, you get one (big!)
star if the information has been made public at all, even if it
is a photo of a scan of a fax of a table -- if it has an open
licence. The you get more stars as you make it progressively more
powerful, easier for people to use.

<table>
  <tr>
    <td class="stars">★</td>

    <td>Available on the web (whatever format) <i>but with an
    open licence, to be Open Data</i></td>
  </tr>

  <tr>
    <td class="stars">★★</td>

    <td>Available as machine-readable structured data (e.g. excel
    instead of image scan of a table)</td>
  </tr>

  <tr>
    <td class="stars">★★★</td>

    <td>as (2) plus non-proprietary format (e.g. CSV instead of
    excel)</td>
  </tr>

  <tr>
    <td class="stars">★★★★</td>

    <td>All the above plus, Use open standards from W3C (RDF and
    SPARQL) to identify things, so that people can point at your
    stuff</td>
  </tr>

  <tr>
    <td class="stars">★★★★★</td>

    <td>All the above, plus: Link your data to other people’s
    data to provide context</td>
  </tr>
</table>

<p>How well does your data do? You can buy <a href="http://www.cafepress.co.uk/w3c_shop.480759174">5 star data
mugs</a>, T-shirts and bumper stickers from the W3C shop at
cafepress: use them to get your colleages and fellows
conference-goers thinking 5 star linked data. (Profits also help
W3C :-).</p>

<p>Now in 2010, people have been pressing me, for governmet data,
to add a new requirement, and that is there should be metadata
about the data itself, and that that metadata should be availble
from a major catalog. Any open dataset (or even datasets which
are not but should be open) can be regisetreed at ckan.net.
Government datasets from the UK and US hsould be regisetred at
data.gov.uk or data.gov respectively. Other copuntries I expect
to develop their own registries. Yes, there should be metadata
about your dataset. That may be the subject of a new note in this
series.</p>

<br/>

<p>Linked data is essential to actually connect the semantic web.
 It is quite easy to do with a little thought, and becomes
second nature.   Various common sense considerations
determine when to make a link and when not to.</p>

<p>The <a href="http://dig.csail.mit.edu/2005/ajar/ajaw/tab">Tabulator</a>
client (running in a suitable browser)  allows you to browse
linked data using the above conventions, and can be used to check
that your linked data works.</p>

<p>References</p>

<p>[Ding2005] Li Ding, et. al.,  <a href="http://ebiquity.umbc.edu/paper/html/id/240/"><span>Tracking RDF Graph Provenance using RDF
Molecules</span></a>, UMBC Tech Report TR-CS-05-06</p>
<hr/>

<h2>Followup</h2>

<p>2006-02 Rob Crowell adapts Dan Connolly's DBView (2004) which
maps SQL data into linked RDF, adding backlinks.</p>

<p>2006-09-05 Chris Bizer et al adapt <a href="http://sites.wiwiss.fu-berlin.de/suhl/bizer/d2r-server/">D2R
Server</a> to provide a linked data view of a database.</p>

<p>2006-10-10 Chris Bizer et al produce the <a href="http://sites.wiwiss.fu-berlin.de/suhl/bizer/ng4j/semwebclient/">Semantic
Web Client Library</a>, "Technically, the library represents the
Semantic Web as a single Jena RDF graph or Jena Model." The code
feteches web documents as needed to answer queries.</p>

<p>2007-01-15 Yves Raimond has produced a <a href="http://moustaki.org/swic/">Semantic Web client for SWI
prolog</a> wit similar functionality.</p>

<p>I have a talk at the 2009 O'Reilly eGovernment 2.0 conference
in Washington DC, talking about "Just a Bag of Chips" @@ref, and
talking about the 5 star scheme. Following that, From InkDroid
blogged summary (and CSS) of my 5 star sceheme adapted here</p>