How the VMF matrix works Godfrey Rust, Rightscom Presentation to VMF seminar

vmf Godfrey Rust, Rightscom: How the VMF matrix works, Nov 2009

How the VMF matrix worksGodfrey Rust, Rightscom

Presentation to VMF seminarBritish Library, LondonNovember 9, 2009

[email protected]

http://cdlr.strath.ac.uk/VMF/index.htm


Everything should be made as simple as possible,

but not simpler.

Albert Einstein



VMF goal

vmf

vocab1

vocab2

vocab3

vocab4

Automatically compute the “best fit” mappings between any two pre-defined vocabularies.



vmf

vocab1

vocab2

vocab3

vocab4

3 stages of VMF project



vmf

vocab1

vocab2

vocab3

vocab4


1. Create the matrix and map vocabularies in (what we’ve mainly done so far).



vmf

vocab1

vocab2

vocab3

vocab4



2. Scheme to scheme mappings out (some first tests of this).



vmf

vocab1

vocab2

vocab3

vocab4



2. Scheme to scheme mappings out (some first tests of this).

3. Establish VMF as an ongoing resource.



RDA (libraries)ONIX (book/serials publishing)DDEX (recorded music)Dublin Core (web metadata)FRBR (libraries)LOM SCORM (education)MARC21 (libraries)DOI (any content)CIDOC CRM (museums and archives)FRAD (libraries)MPEG21 RDD (digital rights)RDA ONIX Framework

Initial schemes to map



Concentrating on:

Resource categoriesResource-to-Resource relatorsResource-to-Party relatorsParty-to-Party relatorsParty categories

Vocabularies

Metadata is increasingly seen to be more about Relators than attributes – FRBR, indecs, CIDOC CRM, RDA all Relationship based.

Always been there – just hidden. 9 of the Dublin Core 15 terms are actually Relators.

Semantic Web/RDF point in the same direction. Distributed data means linking identities – which means Relators.



Approximate figures for the matrix today:

10 schemes53 vocabularies mapped in whole or part500+ concept families8000+ unique terms30,000+ RDF triples

Concentrated on resource-to-party relators (highest overlap, most important verbs) at the outset.

Some further work to do before releasing the matrix publicly.

Matrix stats



The data model of the matrix has to accommodate the data models of all the vocabularies mapped to it, and the relationships between them.

The VMF data model and ontology is based on Rightscom’s COA model, which grew from the <indecs> framework and has much in common with FRBR and CIDOC CRM.

COA also provides the “model of making” – an analysis of the relationship between Works and their Manifestations which is of particular use in library/content provider metadata.

Data model: the challenge

vmf

vocab1

vocab2

vocab3

vocab4



How does VMF work?

Terms are mapped into an ontology (the VMF matrix) built up from “families” of concepts based on verbs.

The matrix can be queried to get the “best fit” match from one term or vocabulary to another.

(Note: the matrix is for machine processing – not for human use, and it is not a cataloguing tool or a standard vocabulary).

vmf

vocab1

vocab2

vocab3

vocab4



Context Create (or Creating Event)

A Concept Family




A Concept Family

Create (to Make something (as a human being)




A Concept Family

ParentMake (to bring something into existence)





A Concept Family



Children Conceive, Originate, Derive, Create Work, Create Perceivable Resource, Create with Tool, Create with Material, Direct, Contribute etc…




A Concept Family


Children Conceive, Originate, Derive, Create Work, Create Perceivable Resource, Create with Tool, Create with Material, Direct, Contribute etc…





A Concept Family



Context Create Agent CreatorResource Creation

A Concept Family



Context CreateAgent CreatorResource Creation

Relator Creator_Creation

A Concept Family




Relator Creator_Creation Creation_Creator

A Concept Family




Relator Creator_Creation Creation_CreatorRelator Create_Creation Creation_CreateRelator Create_Creator Creator_Create

A Concept Family




Relator Creator_Creation Creation_CreatorRelator Create_Creation Creation_CreateRelator Create_Creator Creator_CreateRelator Creator_CreatorRelator Creation_Creation

A Concept Family A Concept Family provides a complete set of terms that describe a type of Event or State (“Context”), always based on a verb.

A Concept Family provides a complete set of terms that describe a type of Event or State (“Context”), always based on a verb.

All relationships (and so most meanings) are based on Events, so this is a good place to start.

All relationships (and so most meanings) are based on Events, so this is a good place to start.

Every term in the VMF matrix is a member of a Concept Family.

Every term in the VMF matrix is a member of a Concept Family.





A Concept Family

How do these terms relate to the vocabularies we are mapping?

How do these terms relate to the vocabularies we are mapping?





A Concept Familyonix:CodeList17Created by






marc21:RelationshipCreator







lom:role_lifecycleauthor








Dc:dc15Creator








Dc:dc15Creator

crm:propertywas created by








Dc:dc15Creator


crm:propertyhas created








crm:classMan-made object

Dc:dc15Creator











frbr:Endeavour

Dc:dc15Creator











frbr:Endeavour

Dc:dc15Creator


rdd:verbsMake










ddex: (nothing)

frbr:Endeavour

Dc:dc15Creator


rda: (nothing)

frad: (nothing)

rdd:verbsMake










ddex: (nothing)

frbr:Endeavour

Dc:dc15Creator



rda: (nothing)

frad: (nothing)

rdd:verbsMake

Every term in a vocabulary maps onto a term in some Concept Family.

Every term in a vocabulary maps onto a term in some Concept Family.



Value of the Concept Family (1)

The Concept Families provide all possible points (“nodes”) in the matrix for vocabulary terms to be mapped.



Context DeriveAgent DeriverResource1 DerivationResource2 Source

Relator Deriver_Derivation Derivation_DeriverRelator Deriver_Source Source_DeriverRelator Derivation_Source Source_DerivationRelator Derivation_Derive Derive_DerivationRelator Deriver_Derive Derive_DeriverRelator Source_Derive Derive_SourceRelator Deriver_DeriverRelator Derivation_DerivationRelator Source_Source

A (bigger) Concept Family



Context DeriveAgent DeriverResource1 DerivationResource2 Source

Relator Deriver_Derivation Derivation_DeriverRelator Deriver_Source Source_DeriverRelator Derivation_Source Source_DerivationRelator Derivation_Derive Derive_DerivationRelator Deriver_Derive Derive_DeriverRelator Source_Derive Derive_SourceRelator Deriver_DeriverRelator Derivation_DerivationRelator Source_Source

Automated naming of Relators (easy to create and find terms).

Automated naming of Relators (easy to create and find terms).

A (bigger) Concept Family




The Concept Family allows large numbers of terms to be generated in the matrix very efficiently.

A single concept will typically produce 15-25 terms (3-5 classes and 8-24 relators).

Less than a quarter of those will directly map onto vocabularies, but most will be used as pathways in computing relationships between different terms.

Because the model is simple in concept, and the process is reasonably flexible and automated, the maintenance of a large and complex ontology is manageable (“as simple as possible…”).




The “family tree” of verbs provide the skeleton of the whole matrix, going from the most general concepts at the top down to the most detailed anyone might need to imagine.

Because of the concept families, the matrix is made up of a number of parallel hierarchies, joined across by the family links.



Contextualize

Exist

ExistStatically

ExistDynamically

Act

Do

DoTransitively

DoIntransitively

DoAsPrincipalAgent

DoAsSupportingAgent

Make

Cause

CauseState

Create

Derive

Adapt

Translate

Compile

Interact

Derive

Use

Perceive

See

Some top level concepts: verbs



Contextualize

Exist

ExistStatically

ExistDynamically

Act

Do

DoTransitively

DoIntransitively

DoAsPrincipalAgent

DoAsSupportingAgent

Make

Cause

CauseState

Create

Derive

Adapt

Translate

Compile

Interact

Derive

Use

Perceive

See

Some top level concepts: verbs



Contextualize

Exist

ExistStatically

ExistDynamically

Act

Do

DoTransitively

DoIntransitively

DoAsPrincipalAgent

DoAsSupportingAgent

Make

Cause

CauseState

Create

Derive

Adapt

Translate

Compile

Interact

Derive

Use

Perceive

See

Some top level concepts: verbs, agents

Agent

Doer

DoerTransitively

DoerIntransitively

DoerAsPrincipalAgent

DoerAsSupportingAgent

Maker

Causer

CauserOfState

Creator

Deriver

Adaptor

Translator

Compiler

Interacter

Deriver

User

Perceiver

Seer



Contextualize

Exist

ExistStatically

ExistDynamically

Act

Do

DoTransitively

DoIntransitively

DoAsPrincipalAgent

DoAsSupportingAgent

Make

Cause

CauseState

Create

Derive

Adapt

Translate

Compile

Interact

Derive

Use

Perceive

See

Some top level concepts: verbs, agents, resources

Agent

Doer

DoerTransitively

DoerIntransitively



Maker

Causer

CauserOfState

Creator

Deriver

Adaptor

Translator

Compiler

Interacter

Deriver

User

Perceiver

Seer

Resource

Existent

StaticResource

DynamicResource

Patient

PatientDoneTo

TransitivePatient

DoAsPrincipalAgent

DoAsSupportingAgent

Output

Consequence

ConsequentState

Creation

Derivation, Source

Adaptatation, SourceOf..

Translator, SourceOf..

Compilation, Component

Input

Derivation, Source

UsedResource

PerceivedResource

SeenResource



Contextualize

Exist

ExistStatically

ExistDynamically

Act

Do

DoTransitively

DoIntransitively

DoAsPrincipalAgent

DoAsSupportingAgent

Make

Cause

CauseState

Create

Derive

Adapt

Translate

Compile

Interact

Derive

Use

Perceive

See

Some top level concepts: verbs, agents, resources

Agent

Doer

DoerTransitively

DoerIntransitively



Maker

Causer

CauserOfState

Creator

Deriver

Adaptor

Translator

Compiler

Interacter

Deriver

User

Perceiver

Seer

Resource

Existent

StaticResource

DynamicResource

Patient

PatientDoneTo

TransitivePatient

DoAsPrincipalAgent

DoAsSupportingAgent

Output

Consequence

ConsequentState

Creation

Derivation, Source

Adaptatation, SourceOf..

Translator, SourceOf..

Compilation, Component

Input

Derivation, Suurce

UsedResource

PerceivedResource

SeenResource

…and of course all the Relators between these classes.

…and of course all the Relators between these classes.



Make From Material

Create Part Of Film

Create Parody Of FRBR Expression

Smell

Moderate Panel

Create Content As Librettist

Conduct Experiment

Design Cover Or Cover Artwork

Take Moving Photograph

Be Human

Perform Music As Primary Performer

See With Microform Reader

Be Changeable Necessarily

Add Ink Outline Or Add Color To Visual

Be HTML

Realize On Mirror Site

Some lower level verbs

However granular or obscure, everything belongs to a Concept Family.

The matrix should be able to include almost any concept at any level of detail – but is extended on a “need to know basis” (the principle of Functional Granularity).

However granular or obscure, everything belongs to a Concept Family.

The matrix should be able to include almost any concept at any level of detail – but is extended on a “need to know basis” (the principle of Functional Granularity).




The relators within a Concept Family, and between different families, create the network of meaning to allow “best fit” mappings to be computed when an exact equivalent is not available.



vmf:Adaptor

vmf:WordsAdaptor

vmf:Translator

vmf:SubtitlesTranslator

vmf:WordsCreator

vmf:TranslatorAndCommentator

vmf:Commentator

Mapping to the matrix



vmf:Adaptor

vmf:WordsAdaptor

vmf:Translator


vmf:WordsCreator


vmf:Commentator


Every term in a vocabulary is given an equivalent term in a VMF concept family…




vmf:Adaptor

vmf:WordsAdaptor

vmf:Translator


vmf:WordsCreator


vmf:Commentator onix:Translated by

onix:Translated withcommentary by

ddex:Translator

Ddex:SubtitlesTranslator






vmf:Adaptor

vmf:WordsAdaptor

vmf:Translator


vmf:WordsCreator




ddex:Translator

ddex:SubtitlesTranslator

Queries can then be used to find the “best fit” mappings between two terms or complete vocabularies.


Mapping scheme to scheme



vmf:Adaptor

vmf:WordsAdaptor

vmf:Translator


vmf:WordsCreator




ddex:Translator

Ddex:SubtitlesTranslator

Mapping scheme to scheme





An early test…

We wrote some simple SPARQL queries which map any one vocabulary to any other, asking it to find for each term:

- exact matches- parents- children- siblings (other children of parents)

Mapping results



Mapping results

This is a sample of vocabulary-to-vocabulary mapping results produced by simple queries from the matrix (some more is shown in the VMF Introduction document).

This is a sample of vocabulary-to-vocabulary mapping results produced by simple queries from the matrix (some more is shown in the VMF Introduction document).

Marc 21 Relationship Onix Code List 17

Actor same as Actor

Adapter Child Dramatized by

Adapter Parent Adapted by

Adapter Sibling Abridged by

Adapter Sibling Other adaptation by

Adapter Sibling Translated by

Architect Parent Designed by

Architect Sibling Cover design or artwork by

Arranger same as Arranged by music

Artist same as By artist

Author Child By author

Author Child By composer

Author Child Software written by

Author Sibling From an idea by

Author of introduction etc Child Introduction by

Author of introduction etc Sibling Commentaries by

Author of introduction etc Sibling Memoir by

Author of introduction etc Sibling Notes by

Author of introduction etc Sibling Summary by

Author of screenplay same as Screenplay by

Cartographer same as Maps by

Commentator same as Commentator



There is rarely a “right” answer to a complete vocabulary mapping – we’re looking for the “best fit”. No mapping can create meaning that is not already in the target scheme – it can only try to find it.Some people will want the decisions made for them, others will want to be given the options so that they can choose.

Do we want “Recommended” mappings for particular vocabulary pairs? Yes – then with what rules, and who agrees the results?

There are many possibilities for refining the ontology structure and query methods, including the addition of conditonal rules that go beyond the basic OWL axioms.

Testing and review also provide the best QA for the ontology (at least 10% of mappings likely to be “less than optimal” at this stage).

VMF is “cooking nicely” but needs time to try different things.

Question: how to get the “best fit”?



Matrix data prep in Excel, extracted automatically to RDF triples (TTL format), reviewed using Protege.

A small amount of inference is required to complete gaps in the super-relator hierarchy.

Current query testing using SPARQL, but the matrix and queries could be put automatically into many logic formalisms and queried with many languages. It will be generally available, so help yourself if you want to experiment.

VMF is about the semantics, not the tools.

Technology



rdfs:subClassOf between a class and its parent(s)rdfs:subPropertyOf between a relator and its parent(s)rdfs:domain the domain class of a relatorrdfs:range the range class of a relator owl:inverseOf reciprocal relators rdf:type membership of classesowl:disjointWith classes with no common membersowl:complementOf disjoint classes that make up a parent conceptowl:intersectionOf classes made by combining two or more conceptsowl:unionOf classes whose members must be one thing or another

Matrix axiomsThese RDFS and OWL axioms provide the core logic

These RDFS and OWL axioms provide the core logic



rdfs:subClassOf between a class and its parent(s)rdfs:subPropertyOf between a relator and its parent(s)rdfs:domain the domain class of a relatorrdfs:range the range class of a relator owl:inverseOf reciprocal relators rdf:type membership of classesowl:disjointWith classes with no common membersowl:complementOf disjoint classes that make up a parent conceptowl:intersectionOf classes made by combining two or more conceptsowl:unionOf classes whose members are one thing or anotherowl:equivalentClass mapped classesowl:equivalentProperty mapped relators

Matrix axioms

vmf:IsResourceInCFvmf:IsContextInCFvmf:IsRelatorInCF

This extra set of “vmf” Relators define the relationships within a Concept Family (CF).

They can be translated into logical rules to allow more complex inference.

This extra set of “vmf” Relators define the relationships within a Concept Family (CF).

They can be translated into logical rules to allow more complex inference.





Matrix rules

The concept families and RDFS/OWL axioms provide a “simple as possible” framework for the matrix, and deliver most of what is needed – but some further logical connections need to be made with rules that go beyond the scope of the basic axioms.

For example…



Displaced attributes

Some terms collapse two or more concept families into one. For example – the “Original Author” of a work.





Author_Work

W Shakespeare

Romeo & Juliet





Author_Work

Adapter_ Source

W ShakespeareJ Smith

Romeo & Juliet





Author_Work

Source_ Adaptation

Adapter_ Source


Romeo & Juliet

Romeo & Juliet – the pop up cartoon version





Author_Work

Source_ Adaptation

Adapter_ Adaptation

Adapter_ Source


Romeo & Juliet






Author_Work

Source_ Adaptation

Adapter_ Adaptation

Adapter_ Source


Romeo & Juliet






Author_Work

Source_ Adaptation

Adapter_ Adaptation

Adapter_ Source


“Original Author”

Romeo & Juliet






Author_Work

Source_ Adaptation

W Shakespeare

SourceAuthor_ DerivationWithSourceAuthor

Romeo & Juliet






Author_Work

Source_ Adaptation

W Shakespeare

SourceAuthor_ DerivationWithSourceAuthor

Romeo & Juliet


A rule is needed in the ontology to show that this one relator actually represents a chain of two existing concept families.



Challenges facing the matrix?

The “lattice” approach is logically robust (a common model for ontology), but it can result in a “combinatory explosion” of concepts with multiple parents. One additional attribute at the top can turn into dozens or hundreds further down – a “combinatory explosion”.

RDF/OWL/SPARQL are relatively easy and accessible – but not as powerful as full FOL (First Order Logic) inference.

No easy remedy – everything has trade-offs. The family approach helps you see both the wood and the trees – but we need to look for best ontology practice to get best results (the matrix can be automatically converted into other formalisms if it is helpful).



Refining the matrix

In next stage we would like to:

- extend mappings and validate them with scheme authorities

- develop and test mapping rules, produce candidate mappings for main vocabulary pairs

- welcome suggestions to extend the semantics or organization of the matrix

- welcome all experiments with inference rules to get “best fit”.

- welcome one or two specific projects which can make practical use of VMF.



The need for authorisation

vmf

vocab1

vocab2

vocab3

vocab4

For example, if vocabulary 2 is to map “as well as possible” to vocabulary 4, then both 2 and 4 should validate their mappings to the matrix.

Mapping should be a two way process – most accurate if both parties agree.

Changes must be maintained - the rate of change is increasing.



Availability

VMF matrix freely available as an RDF/OWL ontology before the end of the year.

Tools? Dependent on next stages. We have used Protégé for viewing and validating.

Further details in the VMF Introduction document.



Uses of VMF?

Authoritative source for public vocabulary mappings (but providing a tool for choice if needed)

Incorporate mappings automatically into schema transformations.

Provide QA or options for scheme/vocabulary developers.

Mapping proprietary schemes? Taxonomies/subject indexes?

No limit to scope in principle – concept family model is generic.

Extend to full schema mapping/transformation?

What other uses or benefits can you see?


How the VMF matrix works Godfrey Rust, Rightscom Presentation to VMF seminar

Documents