The Semantic Web

A Whizz-Bang Guide

In a seminal paper in 2001 he said:

...if properly designed, the Semantic Web can assist the evolution of human knowledge as a whole.

So, WTF is it?

A way for computers to understand what each other mean

I need to set up a meeting with Mark about Amazon

• Which Mark?
• Which Amazon?
• Who am I?
• When are Mark and I both free?
• Which room is available at that time?
• How long?

How do we represent these things?

Mark, Amazon, Me, Rooms, Availability, Times

RDF

Resource Description Framework

General method for conceptual description or modeling of information

Triples

• Everything in RDF is represented as a triple; that is:

  • (subject, predicate, object)
  • Subjects, objects and predicates are shared
  • Almost identical to entity-relation graphs

• Generally, graphs of these things are referred to as Knowledge
• When authoring knowledge, you generally ensure the triples are:

  • (noun, verb, noun)

Predicates

• Luigi is a Man

  • (Luigi, isA, Man)

• Luigi likes volleyball

  • (Luigi, likes, Volleyball)

Graph

Dan is a Man

Dan likes Volleyball

Luigi works at WDS

Dan works at WDS

URIs

• In RDF, all subjects, predicates and objects are URIs

  • (Luigi, isA, Man)

• URIs do not need to be dereferencable

  • If they are more information can be gleaned by accessing the data; IRIs vs URIs
  • "Follow Your Nose" data gathering

• Create your own URIs

  • Harder to integrate with the outside world

• Use existing ontologies or URLs

URIs

• Luigi:

  • http://people.wds.co/luigi
  • https://www.facebook.com/lcardito

• Is A:

  • http://www.w3.org/1999/02/22-rdf-syntax-ns#type
  • usually abbreviated to rdf:type

• Man

  • http://makey-upy.com/#Man
  • But what is a man?

Ontologies

• Ontology is the (philosophical) study of reality and existence.
• An Ontology is a representation of a domain - essentially a schema for data

• Early Days

  • Large unweildy ontologies that attempt to represent everything (e.g. OpenCyc, Freebase)
  • Way too expensive

• Best Practice

  • Microformats - small ontologies for specific domains
  • Map to other ontologies if necessary

• Research

  • Buildling ontologies and mappings automatically from the web (e.g. Yago, dbPedia; Okkam)

Manually Building Ontologies

• A lot like OOP

  • Classes (aka concepts) and subclasses (aka subconcepts)

• Protege ontology editor from Stanford Univeristy

Instances vs Concepts

• A vegetable can be a tuber...

• A tuber can be a potato...

  • Do we stop authoring here?

• A potato can be a Russet Potato

  • Do we stop here?

• A Russet Potato could be a Russet Burbank Potato

  • Do we stop here?

• I have a specific Russet Burbank Potato in my basket (an instance)

Literals

• The object can be a literal, specified by an XSD primitive type:

  • (Luigi, hasName, "Luigi"xsd^^string)
  • (Luigi, hasAge, "28"^^xsd:integer)

RDF Serialization

• RDF-XML is common

• Turtle or ntriples is preferred

  • More readable
  • More concise

<rdf:RDF
    xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
    xmlns:foaf="http://xmlns.com/foaf/0.1/"
    xmlns:facebook="https://www.facebook.com/"
    xmlns:madeUp="http://makey-upy.com/#"
    xmlns:dbp="http://live.dbpedia.org/page/"
    xmlns:wds="http://data.wds.co/#" >
  <rdf:Description rdf:about="http://data.wds.co/#WDS">
    <rdf:type rdf:resource="http://data.wds.co/#Company"/>
  </rdf:Description>
  <rdf:Description rdf:about="http://makey-upy.com/#Man">
    <rdf:type rdf:resource="http://www.w3.org/1999/02/22-rdf-syntax-ns#Class"/>
  </rdf:Description>
  <rdf:Description rdf:about="https://www.facebook.com/lcardito">
    <wds:worksAt rdf:resource="http://data.wds.co/#WDS"/>
    <foaf:interest rdf:resource="http://live.dbpedia.org/page/VolleyBall"/>
    <rdf:type rdf:resource="http://makey-upy.com/#Man"/>
  </rdf:Description>
  <rdf:Description rdf:about="http://data.wds.co/#Company">
    <rdf:type rdf:resource="http://www.w3.org/1999/02/22-rdf-syntax-ns#Class"/>
  </rdf:Description>
  <rdf:Description rdf:about="https://www.facebook.com/dan.potepa">
    <wds:worksAt rdf:resource="http://data.wds.co/#WDS"/>
    <foaf:interest rdf:resource="http://live.dbpedia.org/page/VolleyBall"/>
    <rdf:type rdf:resource="http://makey-upy.com/#Man"/>
  </rdf:Description>
</rdf:RDF>

@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .
@prefix dbp: <http://live.dbpedia.org/page/>.
@prefix madeUp: <http://makey-upy.com/#>.
@prefix facebook: <https://www.facebook.com/>.
@prefix wds: <http://data.wds.co/#>.

wds:Company rdf:type rdf:Class.

wds:WDS rdf:type wds:Company.

madeUp:Man rdf:type rdf:Class.

facebook:lcardito
        rdf:type madeUp:Man;
        foaf:interest dbp:VolleyBall;
        wds:worksAt wds:WDS.

facebook:dan.potepa
        rdf:type madeUp:Man;
        foaf:interest dbp:VolleyBall;
        wds:worksAt wds:WDS.

Knowledge Bases

• A knowledge base (KB) is a database for RDF

• There are many and most can deal with more than 1 million triples (some over billions of triples)

  • 4Store (C++/REST, uses MySQL backend, distributed)
  • H2RDF (Java, uses Hadoop backend, distributed)
  • Virtuoso [Graph Database]
  • Neo4J [Graph Database] (Java/REST, distributed),
  • AllegroGraph [Graph Database] (REST, MongoDB, distributed)

• Java frameworks for dealing with knowledge:

  • JENA
  • OpenRDF/Sesame
  • Blueprints/Tinkerpop (general graph database framework)

Searching

• The RDF search language is SPARQL

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
PREFIX foaf: <http://xmlns.com/foaf/0.1/>
PREFIX dbp: <http://live.dbpedia.org/page/>

SELECT * WHERE { ?s foaf:interest dbp:VolleyBall } LIMIT 10

Inference & Reasonining

• Inference:

  • When asking "what type is..." you end up with a whole heirarchy

• Reasoning

  • Creating ontology based on rules in the ontology
  • OWL has contraints and rules

Linked Data

• Using concepts and instances from other places

  • Like we just did with dbPedia

• How do you find stuff to link to?

  • Search

    • Google; e.g. dbPedia VolleyBall
    • Swoogle - semantic web search engine

  • Linking Services

    • dbPedia Spotlight
    • sameAs.org
    • okkam.org

Why not RDF?

• Lists/Sequences are a pain!
• Large or rare libraries/infrastructure required
• Esoteric; Learning curve steeper than alternatives

JSON-LD

• W3C standard for JSON-Linked-Data
• Easy to use (it's just JSON with a bit extra)
• Easy to deploy (Mongo)
• Compatible with RDF but more features (lists)
• Lightweight libraries in JS, Java, Python + more
• Not optimized for SPARQL

Do we need any of this?

• The researchers use this stuff all the time

• It could give us more structured, more reusable data

  • Data inference is necessary for NLU

However

• Extra cost in creation, training, development
• Potentially new infrastructure

Questions?