wm unit1.6-slides-semantic web-final

Miriam Fernández, [email protected] Knowledge Media Institute, The Open University, UK

Universidad Autónoma de Madrid

Máster Universitario en Investigación e Innovación en Tecnologías de la Información y las Comunicaciones

Escuela Politécnica Superior

Web Mining

Mining semi-structured data

The Semantic Web: From words to meanings

1


Web Mining Máster Universitario en Investigación e Innovación en Tecnologías de la Información y las Comunicaciones

Escuela Politécnica Superior, Universidad Autónoma de Madrid

•  1. Introduction •  Unstructured data vs. structured data •  Data vs. information vs. knowledge •  Motivation

•  2. The Semantic Web •  What is the Semantic Web? •  The Semantic Web weapons: ontologies & representation languages •  The Semantic Web tools: ontology editors & triple stores

2




•  3. Two paths towards the Sematic Web vision •  Metadata embedded in HTML •  Linked Data

-‐  An example •  4. Sematic Web applications

•  Search engines •  Recommendations •  Open government •  Institutions

-‐  The Open University -‐  BBC

3




•  1. Introduction •  Unstructured data vs. structured data •  Data vs. information vs. knowledge •  Motivation

•  2. The Semantic Web •  What is the Semantic Web? •  The Semantic Web weapons: ontologies & representation languages •  The Semantic Web tools: ontology editors & triple stores

4




What is … ?

5




What is … ?

Flexible structure (e.g. tags, taxonomies)

Rigid structure: well defined vocabularies and representation

(e.g. databases)

No structure (e.g. free text)

6




What is … ?

Iván in Italy!

Syntactic interpretation

Semantic interpretation

A picture!

7




Why do we need to add structure? THE CHALLENGE

Syntactic interpretation

Semantic interpretation

A picture!

Is this a person, an object, a place? I don’t understand anything! L

Find the way machines can extract & interpret knowledge!

8




Why is this important? We have entered the Zettabyte era! (a billion terabytes) http://www.emc.com/collateral/demos/microsites/emc-digital-universe-2011/index.htm

9




Why is this important?

•  Information overwhelming •  We need mechanisms to support

-‐  better information search -‐  better information integration -‐  automatic knowledge extraction

•  User generated content is generally unstructured •  Machines can not understand such content!

10





Many pages later I still didn’t find what I was looking for! L The search engine didn’t understood my query

11





Moving to England… looking for a place to live

Available houses

Crime reports in different areas

Amenities

Far too much distributed and unrelated information

12




•  1. Introduction •  Unstructured data vs. structured data •  Data vs. information vs. knowledge •  Why is this important?

•  2. The Semantic Web •  What is the Semantic Web? •  The Semantic Web weapons: ontologies & representation

languages •  The Semantic Web tools: ontology editors & triple stores

13




The Semantic Web “I have a dream for the Web in which computers become capable of analysing all the data on the Web”

"The Semantic Web is an extension of the current Web in which information is given well-defined meaning, better enabling computers and people to work in cooperation."

Tim Berners-Lee, James Hendler, Ora Lassila, The Semantic Web, Scientific American, May 2001

“The Semantic Web is a vision: the idea of having data on the web defined and linked in a way that it can be used by machines not just for display purposes, but for automation, integration and reuse of data across various applications”

http://www.w3.org/2001/sw

14




The Semantic Web

Structured

Unstructured

From the Web of human understandable content The Web of documents and links

To the Web of machine understandable content The Web of objects and relations

15




The Semantic Web weapons: ONTOLOGIES

An ontology is a formal, explicit specification of a shared conceptualization

Formal: machine-readable Explicit: concepts, properties, relations, functions, constraints, axioms are explicitly defined Shared: consensual knowledge Conceptualization: abstract model and simplified vide of some phenomenon in the world that we want to represent

16




The Semantic Web weapons: ONTOLOGIES

How to model the world of Pizzas?

Classes Properties Individuals

ontology knowledge base

Restrictions

Cuatrro Formaggi

hasCountryOrigin

Italy

h>p://owl.cs.manchester.ac.uk/tutorials/protegeowltutorial/resources/ProtegeOWLTutorialP4_v1_3.pdf

17




The Semantic Web weapons: REPRESENTATION AND QUERY LANGUAGES

Assigning unambiguous names (URI)

Expressing and linking data, including metadata (RDF)

Querying data (SPARQL)

Capturing ontologies (OWL)

18




RDF (Resource Description Framework) •  RDF identifies things using Web identifiers (URIs), and describes

resources with properties and property values. •  The triple representation (subject, predicate, object)

<?xml version="1.0"?> <rdf:RDF

xmlns:rdf="h>p://www.w3.org/1999/02/22-‐rdf-‐syntax-‐ns#" Xmlns:epsp=h>p://www.ir.ii.uam.es/personnel#> <rdf:DescripYon

rdf:about="h>p://www.ir.ii.uam.es/personnel/IvanCantador"> <epsp:name>Ivan</epsp:name> <epsp:lastname>Cantador</epsp:lastname> <epsp:naYonality>Spanish</epsp:naYonality> <rdf:type>Assistant Professor</rdf:type>

</rdf:DescripYon> </rdf:RDF>

h>p://www.ir.ii.uam.es/ personnel/IvanCantador

h>p://www.ir.ii.uam.es/personnel/name

Ivan

19




RDFS (RDF Schema)

•  RDFS provides the framework to describe classes and properties. It allows the creation of hierarchies <?xml version="1.0"?> <rdf:RDF

xmlns:rdf="h>p://www.w3.org/1999/02/22-‐rdf-‐syntax-‐ns#" xmlns:rdfs="h>p://www.w3.org/2000/01/rdf-‐schema#" xml:base=” h>p://www.eps.uam.es# <rdfs:Class rdf:ID=”Professor" /> <rdfs:Class rdf:ID=”Assistant Professor">

<rdfs:subClassOf rdf:resource=”#Professor"/> </rdfs:Class> <rdf:Property rdf:ID=“teachesSubject”>

<rdfs:Domain rdf:resource=“#Professor"/> <rdfs:Range rdf:resource=“#Subject"/>

</rdf:Property> </rdf:RDF>

20




OWL (Web Ontology Language) •  OWL is a stronger language with greater machine interpretability than

RDF/RDFS (reasoning support) •  OWL Little / OWL DL / OWL Full

<?xml version="1.0"?> <rdf:RDF

xmlns:rdf="h>p://www.w3.org/1999/02/22-‐rdf-‐syntax-‐ns#" xmlns:rdfs=h>p://www.w3.org/2000/01/rdf-‐schema# xmlns:owl = "h>p://www.w3.org/2002/07/owl#" xml:base=“h>p://www.eps.uam.es# <owl:Class rdf:ID=”Professor”>

<owl:restricYon> <owl:onProperty rdf:resource=“#hasAcademicTitle"/>

<owl:hasValue> PhD^^h>p://www.w3.org/2001/XMLSchema#string</owl:hasValue> <owl:restricYon>

</owl:Class> </rdf:RDF>

Value constraints: owl:allValuesFrom owl:someValuesFrom owl:hasValue Cardinality constraints: owl:cardinality owl:minCardinality owl:maxCardinality ….

Ivan has a PhD -‐> therefore Ivan can be professor!

21




SPARQL (SPARQL Protocol and RDF Query Language) •  SPARQL is a query language for RDF

•  Based on the triple representation (subject, predicate, object) •  SPARQL 1.1 is W3C Recommendation since 21st March 2013

PREFIX epsp: <h>p://www.eps.uam.es> <rdf:RDF

xmlns:rdf="h>p://www.w3.org/1999/02/22-‐rdf-‐syntax-‐ns#" SELECT ?x ?name WHERE { ?x rdf:type epsp:Assistant Professor. ?x epsp:name ?name. ?x epsp:naYonality “Spanish”}

SPARQL allows the construcYon of very powerful queries -‐ Filtering elements -‐ Querying named graphs -‐ Ordering by/ disYnct / reduced / offset / limit -‐ Count / sum / avg / min / max / GroupConcat

Can you Google this?

22




The Semantic Web tools: ONTOLOGY EDITORS

•  An ontology editor lets the creation or manipulation of ontologies

h>p://en.wikipedia.org/wiki/Ontology_editor

23




The Semantic Web tools: TRIPLE STORES •  A triple store lets persistent storage of RDF data and querying via

SPARQL

h>p://www.garshol.priv.no/blog/231.html

24








25




Two paths towards the SW vision: ADDING DATA TO THE WEB

•  Metadata embedded in HTML •  Microformats •  RDFa •  Schema.org •  HTML5

•  Linked Data •  Publish the data online in a standard, web enabled representation

(RDF) •  Make the data web addressable (URI) •  Link with other data

26






•  Search engines •  Social Networks •  Open government •  Institutions


27




Microformats •  Microformats use existing HTML attributes to embed structured data

types in an HTML document •  Expressive power is limited as microformats are only designed to pre-

defined vocabularies •  No interlinking between entities <div class="vcard"> <a class="fn org url" href="h>p://www.eps.uam.es/">EPS</a> <div class="adr"> <span class="street-‐address">Francisco Tomas y Valiente</span > <span class="locality">Madrid</span>, <span class="postal-‐code">28049</span> <span class="country-‐name">Spain</span > </div> </div>

h>p://microformats.org

28




RDFa (Resource Description Framework in Attributes)

•  W3C standard for embedding RDF data in HTML documents •  A set of new HTML attributes ands specs of how to use them

•  RDFa is just a syntax, the publisher has to choose the vocabulary •  RDFa is domain independent

<html> <body vocab="h>p://xmlns.com/foaf/0.1/"> ... <h2 property=”name”>Ivan Cantador</h2> <p><span property=”gender">undefined</span></p> ... </body> </html>

h>p://www.w3.org/TR/xhtml-‐rdfa-‐primer

Web standard since June 2012 h>p://www.w3.org/TR/rdfa-‐core/

29




RDFa (Resource Description Framework in Attributes)

•  Used by Facebook! … among others ;) •  RDF vocabulary to be used in conjunction with RDFa

<html xmlns:og="h>p://opengraphprotocol.org/schema/">

<head>

<Ytle>The Rock (1996)</Ytle>

<meta property="og:Ytle" content="The Rock" />

<meta property="og:type" content="movie" />

<meta property="og:url" content="h>p://www.imdb.com/Ytle/>0117500/" />

<meta property="og:image" content="h>p://ia.media-‐imdb.com/images/rock.jpg" /> …

</head> ...

</html>

30




HTML5 (microdata)

•  HTML introduces inline elements (text-level semantics) to describe specific type of information

<div itemscope itemtype="h>p://data-‐vocabulary.org/Person"> My name is <span itemprop="name">Ivan</span>

Here is my home page: <a href="h>p://arantxa.ii.uam.es/~cantador/"

itemprop="url">h>p://arantxa.ii.uam.es/~cantador/</a> I live in Madrid, and work as an <span itemprop="Ytle">Assistant

Professor</span> at <span itemprop="affiliaYon">UAM</span>. </div>

h>p://dev.w3.org/html5/md-‐LC/#encoding-‐microdata

•  Itemid •  Itemprop •  Itemref •  Itemscope •  itemtype

31




Schema.org •  Provides a collection of vocabularies, a long with the microdata

format, that are recognised by the major search engine providers (de facto standard) <div itemscope itemtype=h>p://schema.org/Movie> <h1 itemprop="name">Pirates of the Carribean: On Stranger Tides (2011)</h1> Director:

<div itemprop="director" itemscope itemtype="h>p://schema.org/Person"> <span itemprop="name">Rob Marshall</span>

</div>

32








33




Linked Data Linked Open Data (September 2011)

34




Linked Data: WHAT IS IT ABOUT?

•  The Web as one huge interconnected data graph •  A Web of interlinked objects

•  Objects described in a standard, web enabled representation (RDF) •  Each RDF document describes the characteristics of a single object,

and links to related objects •  Each object has a unique web address (URI) •  Objects are linked with other objects

-‐  Important: links to the same objects in different datasets (sameAs) •  Guidelines for proper configuration of web servers to serve such

documents •  Vocabularies should be reused!

35




Linked Data: WE TALK ABOUT OBJECTS

Wikipedia

The DBpedia ontology, http://dbpedia.org

36




Linked Data: OBJECTS ARE LINKED!

Moving to England… looking for a place to leave

Available houses Crime reports Amenities

A unique information space

Data Mashups

37




Linked Data •  Advantages:

•  No change to the publishing of the HTML documents •  Data can be published by third party (e.g. DBpedia) •  Data is interlinked

•  Disadvantages: •  Web servers need to be configured to properly handle URIs that identify

concepts instead of documents •  Search engines need to be extended to crawl Linked Data •  Suitable vocabularies/ontologies are not always available •  Data is not always easy to obtain •  Data is not always linked to relevant data •  Data is not always linked to HTML documents

-‐  Web mining: OBTAINING, STRUCTURING AND LINKING DATA

38




Methodological guidelines for publishing Linked Data

h>p://delicias.dia.fi.upm.es/wiki/images/f/f6/04_MGLD.pdf

39




Vocabularies / Schemas / Ontologies

•  Search for suitable vocabularies •  Reuse (if they exist) •  Build new vocabularies based on existing ones (if they do not exist)

40




Obtaining and structuring data

•  From unstructured, semi-structured and structured data sources

Natural Language Processing (NLP) •  Entity extractors •  Relation extractors

TextRunner http://openie.cs.washington.edu YAGO: A Core of Semantic Knowledge Unifying WordNet and Wikipedia, Suchanek et al., WWW 2007.

Textual documents Spreadsheets

NOR2O http://oeg-dev.dia.fi.upm.es/nor2o/#download Cvs2rdf4LOD http://logd.tw.rpi.edu/technology/csv2rdf4lod MappingMaster http://protege.cim3.net/cgi-bin/wiki.pl?MappingMaster

41





DataBases

Whole subarea of research within the SW community RDB2RDF h>p://www.w3.org/2001/sw/rdb2rdf ultrawrap h>p://capsenta.com/ultrawrap RDBToOnto h>p://www.tao-‐project.eu/ researchanddevelopment/ demosanddownloads/RDBToOnto.html

WebAPis

Bizer et al. The RDF Book Mashups: From Web APIs to a Web of Data h>p://www.dvi.uni-‐kl.de/~grimnes/ 2007/06/SFSW07Papers/6.pdf

WebForms

Madhavan et al. Google's Deep-‐Web Crawl. VLDB 2008 h>p://www.cs.cornell.edu/~lucja/ publicaYons/i03.pdf

Bringing the Deep Web to the Surface!

42





HTML Tables

Cafarella et al. WebTables: Exploring the Power of Tables on the Web. VLDB 2008

And many other semi-‐structured data XML, UML, BibText, Tags, … ConverterToRDF h>p://www.w3.org/wiki/ConverterToRdf

Google Refine h>ps://code.google.com/p/google-‐refine And its extension RDF Refine h>p://refine.deri.ie

Other useful tools to obtain and structure data

Google Fusion Tables h>p://support.google.com/fusiontables

43




Linking Data

•  1. Identify suitable datasets as linking targets

•  2. Discover relationships between data items

Data Hub ckan.net

Geonames

DBpedia

http://sws.geonames.org/

6355233

http://mydataset/Madrid

http://dbpedia.org/

resource/Madrid

owl:sameAs owl:sameAs

Silk Framework: h>p://wifo5-‐03.informaYk.uni-‐mannheim.de/bizer/silk Limes: h>p://aksw.org/Projects/LIMES.html

h>p://delicias.dia.fi.upm.es/wiki/images/f/f6/04_MGLD.pdf

44




Linking Data to HTML Documents: THE ANNOTATORS

•  The annotators extract entities (classes / individuals) and relations from the text and link them to object URIs

45








46




Obtaining, structuring and linking data: AN EXAMPLE

Any material about C++?... Ufff... Where to start?!

One unique information

space

BEFORE LINKED DATA

AFTER LINKED DATA

47




Obtaining, structuring and linking data: AN EXAMPLE

Use of Linked Data 1.  Select and extract educational

information from various sources 2.  Reuse well-known vocabularies to

describe and structure the previously extracted data

3.  Link the educational material under a common categorization scheme

Fernandez et al. (2011) Linking Data Across UniversiYes: An Integrated Video Lectures Dataset, 10th InternaYonal SemanYc Web Conference (ISWC 2011), Bonn

48




Specification: SELECTING & ANALYSING DATA •  VideoLectures.net

Web scrapper Study the structure of HTML pages and use a HTML parser to extract the data of interest

49




Specification: SELECTING & ANALYSING DATA

•  YouTube University Channels <entry gd:etag='W/"DkADSH47eCp7ImA9WhZWFEg."'> <id>tag:youtube.com,2008:video:zZCaHSW88Ts</id> <published>2011-02-18T11:41:08.000Z</published> <updated>2011-05-15T10:19:39.000Z</updated> <category scheme='http://gdata.youtube.com/schemas/2007/categories.cat' term='Education' label='Education'/> <category scheme='http://gdata.youtube.com/schemas/2007/keywords.cat' term='Dr Barry Cooper'/> <title>Intro to Professional Practice (Children & Families)</title> <author> … </author> <media:description … </media:description> <media:keywords>…</media:keywords> <media:thumbnail …/> <yt:duration seconds='399'/> <content …'/>

YouTube API: select the the fields of interest

50




Specification: SELECTING & ANALYSING DATA

•  OU Podcasts (data.open.ac.uk)

Already structured data: SPARQL query

51




Modeling: REUSING VOCABULARIES

•  Dublin Core •  http://dublincore.org/documents/dcmi-terms (dcterms)

•  FOAF •  http://xmlns.com/foaf/spec (foaf)

•  The W3C ontology for media resources •  http://www.w3.org/TR/mediaont-10 (ma)

•  The Media Vocabulary •  http://payswarm.com/vocabs/media (media)

•  The Nice Tag Ontology •  http://ns.inria.fr/nicetag/2010/09/09/voc.html (nt)

Vocabularies selected to describe the data

I describe tags

We describe media resources

I describe people and relations

I describe educational

material

52




Modeling: REUSING VOCABULARIES

The selected base URI is: http://linkeduniversities.org 1.  VideoLectures objects are represented as media:Recording,

authors are represented as foaf:Person 2.  The video title is duplicated in the properties rdfs:label &

dcterms:title

4.  The set of tags and categories associated to a video is represented by the nt:isRelatedTo property

5.  The assigned classification in the unified search space is represented by the dcterms:subject property

53




RDF generation: CREATING OBJECTS

http://linkeduniversities.org/video/CarnegieMellonU/youtube/B135229F3706D215

rdf:type media:Recording media:download http://www.youtube.com/watch?v=TOTuStPIeFc&feature=youtube_gdata_player dcterms:title CMU Football Engineering Summer 2008 Video rdfs:label CMU Football Engineering Summer 2008 Video dcterms:description Football […]Summer 2008 Video

foaf:thumbnail http://i.ytimg.com/vi/TOTuStPIeFc/3.jpg media:duration 155 dcterms:isPart http://linkeduniversities.org/video/CarnegieMellonU/youtube/playlist/B135229F37

ma:publisher http://linkeduniversities.org/video/CarnegieMellonU/youtube/user/footballtracking dcterms:published 2011-06-03T23:23:53.262Z

nt:isRelatedTo http://linkeduniversities.org/video/CarnegieMellonU/tag/sports nt:isRelatedTo http://linkeduniversities.org/video/CarnegieMellonU/tag/football dcterms:subject http://dmoz.org/Sports/Football/Rugby_Union dcterms:subject http://linkeduniversities.org/video/CarnegieMellonU/dmoz/Sports/Football/Rugby_Union

Type Provenance ID

54




Links Generation: SELECTING THE LINKING POINTS •  A common categorization scheme:

1.  The International Press Telecommunications Council (IPTC) http://www.iptc.org/site/NewsCodes/

2.  Library of Congress Subject Headings http://id.loc.gov/authorities/about.html

3.  The Open Directory Project (DMOZ) http://www.dmoz.org/rdf.html

4.  DBpedia Categories http://dbpedia.org/About

55




Links Generation: EXPLOITING THE ANNOTATORS (1) Extract the information from the video lecture

(2) Generate an textual document (3) Provide the document to the Textwise classification service

Reference/Knowledge_Management (id=495), weight=0.71

(4) Obtain the ODP document classification

56




PUBLICATION •  New LD educational dataset

•  More than 14,000 video lectures from 27 institutions

57








58




Search engines: BETTER DATA VISUALISATION

•  2007 Yahoo! Presented Search Monkey

•  2009 Google announced Rich Snippets

If search engines could understand the data inside the HTML pages they would display such data in better ways: pictures, dates, prices, ratings, etc.

59




Search engines: BETTER CONTENT RETRIEVAL

2012 Google announced Knowledge Graph http://googleblog.blogspot.co.uk/2012/05/ introducing-knowledge-graph-things-not.html

60




Search engines: NOT ALL IS ABOUT DOCUMENTS •  Search engines oriented to concept-search vs. document search

FreeBase: In July 2010 Google bought Metaweb, the company behind freebase

Powerset: Bought by Microsoft in 2008

Wolfram Alpha: Active since 2009

61








62




Recommendations: FACEBOOK GRAPH SEARCH Links among people, places, music, and other objects can lead to better content recommendation

63








64




Open Government •  Open government

•  Data.gov •  Data.gov.uk •  Many others…

Research Funding Explorer

•  Linking data lets more complex queries and deeper data analysis

•  Organizations can now easily link to public government data

•  Favour transparency

65








66




Exposed as linked data, our data interlink with each other and the external world: become part of the “global data space” on the Web

OU public data sit in different systems – hard to discover, obtain, integrate by users

OU public data sit in different systems – hard to discover, obtain, integrate by users

Exposed as linked data, our data interlink with each other and the external world: become part of the “global data space” on the Web

The Open University

ORO

Archive of Course Material

Library’s Catalogue Of Digital Content

OpenLearn Content

A/V Material Podcasts iTunesU

Data from Research Outputs

DBLP

RAE

data.gov.uk

67




The Open University •  The OU has been the first University to expose its data as linked data:

http://data.open.ac.uk •  Now widely recognized as a critical step forward for the HE sector in

the UK (and worldwide) •  Favor transparency and reuse of data, both externally and internally •  Reduces cost of dealing with our own public data: integration and reuse by

design •  Enable both new kinds of applications, and to make the ones that are already

feasible more cost effective

68




The Open University •  Linking Open Learn with relevant Podcasts

69




The Open University •  Exploring research communities

70








71




BBC: British Broadcasting Corporation

•  Historically when developing websites… •  Hand-crafted customized sites •  Only some programs could be covered

-‐  Shame considering they broadcast between 1,000 and 1,500 programs/day

•  All those sites were developed in isolation -  Often not maintained -  Often not persistent

Using Linked Data to build Websites more productively http://www.slideshare.net/metade/linked-data-on-the-bbc

72





•  The Linked Data Website construction model Domain-driven design •  Identify the domain objects

and relations •  Check the domain model with

users •  Design your schema •  Design your URIs •  Define data you need to build

each of your pages •  Automatically build up your

HTML pages by querying these data and applying CSS layouts

73





•  Starting with the WorldCup 2010… and now… •  Programs •  Music •  Artists •  Animal life •  …..

74




Thx! J

•  Questions, doubts?

wm unit1.6-slides-semantic web-final

Education

ivan cantador