大象传媒

A skim-read introduction to linked data

The internet

The internet ≠ the web

The web

https://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

...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...

On SEO (sorry)

• in the days of Alta Vista and Yahoo! mk 1 search was all about documents
• metatags...
• ...and keyword density
• Google changed the game with PageRank by making search...
• ...less about keyword density...
• ...and more about link density

One problem with the web

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

We need to get from this...

..to this

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

...but machines can't

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

RDF

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

More examples

• <Yves> <was born in> <France>
• <France> <is part of> <the EU>
• <Michael> <was born in> <the UK>
• <The UK> <is part of> <the EU>
• <Yves> <likes> <open data>
• <Michael> <likes> <open data>

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

On 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 Yves...

• ...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

Content negotiation - what I'm given

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

Content negotiation - what I'm not given

An honorary mention for One Web

• The Web is designed as a universal space. Its universality is its most important facet. I spend many hours giving talks just to emphasize this point. The success of the Web stems from its universality as do most of the architectural constraints. #timbl
• So one URI per resource; many representations per resource
• Mobile demands a different representation...
• ...but not a different resource...
• ...and not a different URI
• so no mobile. or .mobi etc

Linked data

• is sort of a second pass at the semantic web
• still uses rdf (although that can be contentious)
• as the name implies puts the emphasis back on links
• is about things not documents

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.

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

• /programmes/b006m86d is an information resource about Eastenders
• you can ask for and get it as html or xhtml-mp or rdf...
• but it is not Eastenders
• is an information resource about Yves
• but it is not Yves

Non-information resources

• why do we need the distinction?
• is saying represents Yves good enough?
• Yves was not created when was created
• The person who created (Yves) did not create Yves
• I am not friends with
• The people who created /programmes/b006m86d did not create "Eastenders"

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

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

Given 2 websites...

The microformats approach: XFN @rel

The linked data approach: 2 new non-information resources...

...each with a homepage...

...tying together the 2 people, not the 2 documents

Hashes

• you can give the non-information resource a hash URI
• /programmes/b006m86d is the URI for Eastenders
• if you attempt to dereference /programmes/b006m86d it can't be sent down the wires...
• ...but there's no need for a 303...
• .../programmes/b006m86d is content negotiated to either HTML or RDF depending on what you accept
• and your user agent looks for a #programme inside it

Slashes

• or you can give the non-information resource a completely different path
• is the URI for Holborn
• if you attempt to dereference it can't be sent down the wires...
• ...so it 303s (see other)...
• to - an information resource about Holborn...
• ...which is then content negotiated to either HTML or RDF depending on what you accept

URIs in pictures: slash + 303 + conneg

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

URIs in pictures: hash + conneg

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

URIs in pictures: RDFa

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

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

• The 大象传媒 knows when it's played a record by The Fall
• knows all the records The Fall have ever released
• knows The Fall are from Salford...
• knows New Order are from Salford...
• The 大象传媒 knows when it's played a record by New Order
• etc

owl:sameAs

• allows us to declare that non-information resources across the web are the same thing
• so the 大象传媒's The Fall is the same as...
• ...the MusicBrainz The Fall is the same as...
• ...the DBpedia The Fall is the same as...
• ...the last.fm The Fall is the same as...
• ...etc

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

...and this is not Hamlet

It is...

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

FRBR

• is a library science thing
• it stands for Functional Requirements for Bibliographic Records
• it's used to break down the book example into works, expressions, manifestations and items

FRBR can also be used to describe music

Once you've minted a URI for a non-information resource

• anybody else can make claims about that resource (including sameAs)
• the claims may be true... or not
• we still need trust mechanisms
• [and we should be particularly careful when minting URIs for people]

There are vocabularies available for

• erm,
• etc

And if an ontology doesn't exist

• you can make one

Linking open data

• A community driven project...
• ...to take data sets published under a liberal licence...
• ...and express them as linked data.

The LOD cloud - baby steps

The LOD cloud - today

Fin!

Questions?