An introduction to linked data

Lets start with some egg sucking

The internet

...provides a means to connect machines

The internet ≠ the web

World Wide Web

The web

...provides a means to connect documents

The web = the internet

The web = the internet + links

The web = the internet + links + documents

...or....

The web = the internet + http + html

Web standards

http://en.wikipedia.org/wiki/Web_standards#Common_usage

When a web site or web page is described as complying with web standards, it usually means that the site or page has valid or nearly valid HTML, CSS and JavaScript. The HTML should also meet accessibility and semantic guidelines.

http://en.wikipedia.org/wiki/Web_standards#Common_usage

When a web site or web page is described as complying with web standards, it usually means that the site or page has valid or nearly valid HTML, CSS and JavaScript. The HTML should also meet accessibility and semantic guidelines.

We tend to obsess on the documents:

"Semantic" HTML
HTML, HTML2, HTML3, HTML4, XHTML, HTML5...
DOM
?Flash?
ufs
CSS
document accessibility
validation

...at the expense of the links:

HTTP
URIs

The trouble is...

...HTML has always been...

...a bit of a hack

Everything that's good about the web comes from links

wikis
blog permalinks
citations
sharing
social bookmarking
twitter
socialness in general...

If you can point at something you can talk about it and share it

The web = the internet + http + html

Magazines are made of pages....

...websites are made of links

One problem with the web

people aren't really interested in documents
they're interested in things

We need to get from this...

a web of documents

..to this

a web of things

The other problem with the web...

...people can parse documents and extract meaning...

meaning

...but machines can't

no meaning

We need to help machines to understand the web...

...so machines can help us to understand things

The semantic web

Mk 1

an attempt to make documents that machines could understand
using RDF
before we go any further...

RDF

is a way to model data
is not a serialisation
can be serialised as XML
or inside HTML (as RDFa)
or N3
or Turtle
or even json

The RDF data model

RDF

is based on triples
subject, predicate, object
<The sky> <has the colour> <blue>

More examples

<Bernard Sumner> <was a member of> <New Order>
<Bernard Sumner> <was born in> <Salford>
<Tony Wilson> <was born in> <Salford>
<Tony Wilson> <founded> <Factory Records>
<New Order> <were signed to> <Factory Records>
<New Order> <were directors of> <The Hacienda>
<Tony Wilson> <was a director of> <The Hacienda>

Triples combine to form a graph...

A graph of the relationships in the previous slide

...which is fundamentally different to the set based approach of relational databases

A venn diagram of the relationships in the previous slide

Graphs are web like - so easily expandable and scaleable

An expanded graph

One night in Manchester - birth of Factory Records

An expanded graph

So what happened

lots of people made interlinked foaf files
but there was a shortage of vocabularies to describe other things
so the foaf files only really linked to other foaf files
it all became a little document-y
and less link-y

Semweb mk 1 = the internet + http + rdf

REST

described by Roy Fielding
about the proper use of HTTP (post, get, put, delete)
separated the resource (abstract) from...
the representation (concrete-ish)

If I ask for a document about Factory Benelux...

...my user agent (browser) sends additional information...
...about what I accept in the form of accept headers
I may accept French but prefer English
I may accept JSON but prefer RDF
I may accept full fat xhtml but prefer xhtml-mp

The key is

a URI identifies a resource
not a representation
I ask for a resource - the server returns the best possible representation

Content negotiation - what I want / what I accept

I'd like this resource about Factory Benelux, I speak English but I can just about get by in French and I'd like it for my mobile, please

Content negotiation - what I'm given

Can't do you English but I've got French and can send as xhtml-mp. Here you go

Not always successful

the resource may exist
but with no representation that matches your accept headers
406 - Not acceptable
My favourite HTTP code :-)

Content negotiation - what I want / what I accept

I'd like this resource about Factory Benelux, I speak English but I can just about get by in German and I'd like it for my mobile, please

Content negotiation - what I'm not given

I've got that resource but can only do French. 406

Back to linked data

Linked data

is sort of a second pass at the semantic web
still uses rdf
as the name implies puts the emphasis back on links
is about things not documents

Linked data = the internet + http + rdf

Linked data = web standards

Design issues for linked data

Use URIs as names for things
Use HTTP URIs so that people can look up those names
When someone looks up a URI, provide useful information, using the standards (RDF, SPARQL)
Include links to other URIs. so that they can discover more things.
http://www.w3.org/DesignIssues/LinkedData.html

Use URIs as names for things (my emphasis)

separating resource from representation (REST) gets us so far
but it's still about getting resources as information documents...
...and we still think primarily about information resources

The map is not the territory

http://www.factoryrecords.net/ is an information resource about Factory Records
it might be available as HTML or json or RDF
but it is not Factory Records
http://moustaki.org/ is an information resource about Yves
but it is not Yves

Non-information resources

why do we need the distinction?
is saying http://moustaki.org/ represents Yves good enough?
Yves was not created when http://moustaki.org/ was created
The person who created http://moustaki.org/ (Yves) did not create Yves
I am not friends with http://moustaki.org/
And the people who created http://www.factoryrecords.net/ did not create Factory Records

We want to be able to make different claims about the thing and the document about the thing

So we need URIs for non-information resources - stuff that you can't send down wires

What happens if someone asks for a non-information resource?

I'd like Factory Records, and by the way I speak English but I can just about get by in French

What happens if someone asks for a non-information resource?

Yves will not fit down the wires but (303) I can give you some information about him in English

Designing URIs for non-information resources

Slash URIs

You can give the non-information resource a completely different path
http://dbpedia.org/resource/New_Order is the DBpedia URI for New Order
if you attempt to dereference http://dbpedia.org/resource/New_Order it can't be sent down the wires...
...so it 303s (see other)...
to http://dbpedia.org/page/New_Order - an information resource about New Order...
...which is then content negotiated to either HTML or RDF depending on what you accept

Slash URIs in pictures: slash + 303 + conneg

slash uris

You need to be able to configure your server for 303s and content negotiation.

Hash URIs

or you can give the non-information resource a hash URI
http://www.bbc.co.uk/music/artists/f1106b17-dcbb-45f6-b938-199ccfab50cc#artist is the BBC URI for New Order
if you attempt to dereference http://www.bbc.co.uk/music/artists/f1106b17-dcbb-45f6-b938-199ccfab50cc#artist it can't be sent down the wires...
...but there's no need for a 303...
...http://www.bbc.co.uk/music/artists/f1106b17-dcbb-45f6-b938-199ccfab50cc is content negotiated to either HTML or RDF depending on what you accept
and your user agent looks for a #artist inside it

Hash URIs in pictures: hash + conneg

hash uris

Cheaper setup - no need to set up for 303s although you still need content negotiation. Fewer round trips to server.

RDFa URIs in pictures: hash only

RDFa uris

Cheapest setup - no need to set up for 303s or content negotiation.

So, what's the point?

Different people know (or claim to know) different things about the same topic

The BBC knows when it's played a record by New Order
MusicBrainz knows all the records New Order have ever released
DBpedia knows New Order are from Salford...
DBpedia knows The Fall are from Salford...
The BBC knows when it's played a record by The Fall
etc

Linked data is a web-scale database

A special mention for owl:sameAs

owl:sameAs

allows us to declare that non-information resources across the web are the same thing
so the BBC's New Order is the same as...
...the MusicBrainz New Order is the same as...
...the DBpedia New Order is the same as...
...the last.fm New Order is the same as...
...etc

So if we say...

http://www.bbc.co.uk/music/artists/d5da1841-9bc8-4813-9f89-11098090148e#artist
<owl:sameAs>
http://dbpedia.org/resource/New_Order
we can combine knowledge that's currently scattered across the web...
...and construct a knowledge that's greater than the sum of its parts

When sameAs goes wrong

Language is ambiguous
We often use the same labels to refer to different things in different contexts

An example stolen from Tom Heath

I have in my house a bottle of Bowmore single malt whisky...
...which is one bottle of one vintage (Bowmore 18 year old single malt)...
...which is one vintage of one brand (Bowmore single malt)...
...which is distilled by the Bowmore Distillery
The Distillery, the brand, the vintage and my bottle are not the same thing...
...but we use the same label

When we declare sameAs we need to be careful

If I said my bottle of Bowmore whisky was the sameAs the Bowmore Distillery and someone else said the Bowmore Distillery was created in 1779...
...then by implication my bottle would be have been created in 1779...
...which would be quite a vintage

When using sameAs you need to decide

is this the same thing in a different context
or does the different context make it a different thing
which can get tricky

Ceci n'est pas une pipe...

...and this is not Hamlet

Photo of book cover of Hamlet

It is...

a photograph of...
...one copy of...
...one edition of...
...one version of...
...Hamlet