Linked Data for Librarians

Tim Berners-Lee proposed
4 principles to publish Linked Data.

1. Use URIs as names for things.
2. Use HTTP URIs so people
   can look up those names.
3. When someone looks up a URI,
   provide useful information using the standards.
4. Include links to other things,
   so people can discover more.

(Non-)information resources
should be uniquely identifiable.

• Pointing to resources with a common name
  or description is often ambiguous.
  • Washington (city? state? president?)
• Reuse or mint a URL for them.
  • http://dbpedia.org/resource/George_Washington
• Especially machines need unambiguous identifiers.

The URL is part of a broader family
of technologies related to identification.

• URL – Uniform Resource Locator
  • unique identification and location of resources
  • mailto:ruben.verborgh@ugent.be
• URN – Uniform Resource Name
  • location-independent resource identifier
  • urn:isbn:0-83891251-6
• URI – Uniform Resource Identifier
  • union of URLs and URNs

The broadest family is IRI,
which supports non-ASCII characters.

• Not all characters are allowed in a URI.
  • Percent-encode non-ASCII chars.
    • é becomes %C3%A9
  • Percent-encode chars with other meaning.
    • ? becomes %3F
• IRI – Internationalized Resource Identifier
  • Non-ASCII chars don’t need to be encoded.
  • Chars with other meaning still need encoding.

An HTTP URL identifies and locates a resource anywhere in the universe.

• A string is a unique identifier if
  at most one entity corresponds to it.
  • A national number uniquely identifies a person,
    but does not allow locating him or her.
• A string is a unique locator if
  at most one location corresponds to it.
  • A street address uniquely identifies a location,
    but does not allow to identify a specific person.

Using HTTP URLs ensures that
anybody can look up the resource.

• An HTTP URI of a resource can be dereferenced:
  use an HTTP client to retrieve a representation.
  • Information resources result in a representation.
  • Non-information resources result in a 303 redirect.
• This relies on the double role of an HTTP URI
  as identifier and locator.
• Dereferencing is a core principle of Linked Data.
  • If you don’t know something, look it up.

Dereferencing a URI should lead to
useful information about that resource.

• “Useful” means the information is available
  using standard technologies.
  • Tim Berners-Lee mentions RDF and SPARQL.
• “Useful” also means the information provides
  explanations and/or context for the resource.
• Define the resource in terms of concepts
  the client already knows or can look up.

By including links to other resources,
we create a Web of Data.

• Links connect a resource to known concepts.
  • George Washington was born in British America
• Links give meaning to data.
  • These temperatures are measured in degrees Celsius.
• Links allow exploration of related data.
  • Find more by the same author.

An immense amount of Linked Data
is available on the Web for reuse.

• On the structural level, hundreds of vocabularies
  can provide building blocks to model your data.
  • They provide properties and classes to reuse.
  • Although not always counted as Linked Data
    because of their small size, most follow all 4 principles.
• On the content level, thousands of datasets
  provide identifiers and data of individuals.
  • Strive to reuse identifiers rather than to mint new ones.

No Linked Data set is ever complete.
We make the open-world assumption.

• Relational databases use highly rigid structures.
  They strive for complete data.
  • A NULL value can signal missing data,
    but this is typically an undesired situation.
• With Linked Data, no source has all of the truth.
  Other sources might have more data on a subject.
  • The absence of a fact does not imply its falsehood.
  • A fact has 2 possible states: true and unknown.

Several vocabularies are used frequently
across different datasets.

• modeling vocabularies
  • RDFS
  • OWL
  • SKOS
• general-purpose vocabularies
  • Dublin Core
  • …
• concept-specific vocabularies
  • Schema.org
  • DBpedia ontology
  • …

Find the one you need at Linked Open Vocabularies.

The Dublin Core terms are a set of
15 common metadata properties.

• Each property is generic,
  and hence applicable in many cases.
  • title
  • description
  • date
  • creator
  • …
• Many applications use the Dublin Core terms.
  • good interoperability of high-level semantics

Schema.org is a single vocabulary
that covers many different fields.

• Created and maintained by major search engines,
  it mainly provides discovery data for search.
• Its concepts are defined rather loosely.
  • This makes it flexible to use for developers.
  • Machines cannot derive much knowledge from it.
• Schema.org is manually curated
  and open for extension.

Billions of Linked Data facts are
on the Web with an open license.

• The most well-known dataset is DBpedia.
  • Data is extracted automatically from Wikipedia.
  • Like Wikipedia, it exists in several different languages.
  • Its quality is acceptable for many queries.
• Wikidata is a manually curated alternative.
  • It has its own data model on top of RDF.
  • It grows fast and might overtake DBpedia.
• You can find many other datasets on Datahub.

RDF Schema is an RDF vocabulary
to model other RDF vocabularies.

• RDFS defines classes, properties, and datatypes
  that are used to define vocabularies.
  • The vocabulary has a human-readable specification.
  • It also has a machine-readable vocabularies,
    which define RDFS in terms of RDFS.
• RDFS defines concepts in two namespaces.
  • rdf: https://www.w3.org/1999/02/22-rdf-syntax-ns#
  • rdfs: https://www.w3.org/2000/01/rdf-schema#

Practitioners in the RDF world often
refer to vocabularies as ontologies.

• Strictly speaking, a vocabulary is a set of words;
  an ontology a set of concepts and their relations.
• The W3C states that there is no clear division.

  vocabulary

  usually in less formal contexts

  ontology

  usually in more complex and/or formal contexts

The Web Ontology Language (OWL) provides concepts for ontologies.

• RDFS captures basic ontological relations,
  but lacks several common and important concepts.
  • cardinality restrictions on properties
  • inverse, symmetric, and transitive properties
  • equality and disjointness
  • …
• OWL extends RDFS with advanced concepts.
  • RDFS and OWL are used side by side.

SPARQL Protocol And Query Language: query and update RDF datasets.

• SPARQL is a query language.
  • Select specific data from an RDF dataset.
  • Insert, change, or delete data in an RDF dataset.
• The SPARQL protocol is a Web API definition
  for querying in the SPARQL language over HTTP.
  • A SPARQL endpoint executes SPARQL queries by clients
    through HTTP, and replies with their results.

The SPARQL language defines
forms a query can take.

There are currently 4 read-only query forms:

SELECT

find values that satisfy conditions

CONSTRUCT

create triples that satisfy conditions

ASK

check whether data exists

DESCRIBE

show information about a resource

The main unit of a SPARQL query
is a Basic Graph Pattern (BGP).

• A BGP is a set of triple patterns.
  • Their syntax is a superset of Turtle.
• A triple pattern is a triple in which
  each of the components can be a variable.
  • Variables start with a question mark (?name).
• A SPARQL query engine finds solution mappings.
  • Variables and blank nodes are mapped to URIs,
    blank nodes, or literals according to dataset triples.

This query searches DBpedia for artists influenced by Picasso.

PREFIX dbr: <http://dbpedia.org/resource/>
PREFIX dbo: <http://dbpedia.org/ontology/>
SELECT DISTINCT ?person ?personLabel WHERE {
  ?person a dbo:Artist.
  ?person foaf:name ?personLabel.
  ?person dbo:influencedBy dbr:Pablo_Picasso.
}

Here is the live result of that query.

This query searches Wikidata for artists influenced by Picasso.

PREFIX wd: <http://www.wikidata.org/entity/>
PREFIX wdt: <http://www.wikidata.org/prop/direct/>
SELECT DISTINCT ?person ?personLabel WHERE {
  ?person wdt:P106/wdt:P279* wd:Q483501.
  ?person wdt:P737 wd:Q5593.
  SERVICE wikibase:label { bd:serviceParam wikibase:language "en" }
}

Here is the live result of that query.

Why are the results
and the queries different?

• DBpedia and Wikidata contain different data.
  • So far, so good. This was expected.
• DBpedia and Wikidata use different ontologies.
  • Ideally, the same SPARQL queries would suffice.
  • …and they can, when ontologies link to each other.
• In practice, some bridging is still needed.
  • Semantic Web reasoning can bridge the gap,
    but is often not enabled on query endpoints.

Heterogeneity exists on multiple levels
across metadata collections.

• Heterogeneity exists on the data level.
  • We can choose our own vocabularies.
  • How do we ensure they align?
• Heterogeneity exists on the interface level.
  • We can choose how consumers can query our data.
  • How can clients consume multiple datasets easily?

Heterogeneity is our best friend
and our largest enemy.

• Anybody on the Web is free
  to publish however they want.
  • This works great for people—sometimes.
  • It often doesn’t work great for machines.
• Standardization helps us align.
  • delicate balance between flexibility and interoperability

Standardization and agreement
have provided us with foundations.

• the Semantic Web family of standards
  • RDF
  • SPARQL
  • RDFS
  • OWL
  • …
• ontologies and vocabularies
  • Dublin Core
  • DBpedia ontology
  • Wikidata ontology
  • Schema.org
  • …
• Web APIs not yet
  • Linked Data Platform
  • OAI-ORE

The current level of standardization
still leaves some areas uncovered.

• vocabulary usage
  • examples and guidance
• vocabulary agreement
  • the right terms for the right clients
• Web APIs
  • stop reinventing the wheel

Which vocabularies should we use
to describe our metadata, and how?

• We need to develop examples and guidance.
  • vocabulary usage
  • URL strategy
  • …
• Reasoning can fill vocabulary gaps.
  • You can apply this to your own metadata.
• We can never cover all vocabularies.
  • Negotiate content through profiles.

Web APIs are the Achilles’ heel
of interoperability on the Web.

• Shall we all have our SPARQL endpoints?
  • Who is going to pay?
• Shall we all support the Linked Data Platform?
  • That doesn’t solve querying…
• Shall we all have our own custom APIs?
  • That’s not a sustainable way.

See more in the REST module.