1. Metadata for the Web From Discovery to Description
CS 502 –
Carl Lagoze – Cornell University
Cornell CS 502

2. The fifteen Dublin Core Elements
Cornell CS 502

3. A Pidgin for Digital Tourists
Metadata is language
Dublin Core is a small and simple language -- a pidgin -- for finding resources across domains.
Speakers of different languages naturally "pidginize" to communicate
E.g., tourists using simple phrases to order beer ("zwei Bier bitte" "dva pivo" "biru o san bai"...)
We are all "tourists" on the global Internet.
Cornell CS 502

4. What is the Dublin Core (1)
A simple set of properties to support resource discovery on the web (fuzzy search buckets)?
Domain
Independent
view
Cornell CS 502

5. An extensible ontology for resource desciption?
What is Dublin Core (2)?
An extensible ontology for resource desciption?
Greater Functionality & Cost
Cornell CS 502

6. What is the Dublin Core (3)?
A cross-domain switchboard for interoperable metadata?
Dublin
Core
MARC
INDECS
IMS
- projections to application-specific metadata vocabularies
Switchboard
Cornell CS 502

7. Dublin Core Qualifiers
From fuzzy buckets to more specific description
Model of “graceful degradation”
Support both simplicity and specificity
Intra-domain and inter-domain semantics
Cornell CS 502

8. Varieties of qualifiers: Element Refinements
Make the meaning of an element narrower or more specific.
Narrowing implies an is a relationship
a "date created“ is a "date“
an "is part of relation“ is a "relation“
If your software does not understand the qualifier, you can safely ignore it.
Cornell CS 502

9. Varieties of Qualifiers: Value Encoding Schemes
Says that the value is
a term from a controlled vocabulary (e.g., Library of Congress Subject Headings)
a string formatted in a standard way (e.g., " " means May 3, not February 5)
Even if a scheme is not known by software, the value should be "appropriate" and usable for resource discovery.
Cornell CS 502

10. A Grammar of Dublin Core
By design not as subtle as mother tongues, but easy to learn and extremely useful in practice
Pidgins: small vocabularies (Dublin Core: fifteen special nouns and lots of optional adjectives)
Simple grammars: sentences (statements) follow a simple fixed pattern...
Cornell CS 502

11. Example Dublin Core statements
Resource has Title 'Grammar of Dublin Core'.
Resource has Creator 'Tom Baker'.
Resource has Subject 'Metadata'.
Resource has Relation
Cornell CS 502

12. Resource has property X implied verb one of 15 properties
property value
(an appropriate
literal)
DC:Creator
DC:Title
DC:Subject
DC:Date...
implied
subject
Resource
has
property X
Cornell CS 502

13. Resource has property X implied verb one of 15 properties
property value
(an appropriate
literal)
DC:Creator
DC:Title
DC:Subject
DC:Date...
implied
subject
Resource
has
property X
qualifiers
(adjectives)
[optional qualifier]
[optional qualifier]
Cornell CS 502

14. Resource has Subject "Languages -- Grammar" Resource has Date
LCSH
Resource
has
Date
" "
ISO8601
Revised
Cornell CS 502

15. Dumb-Down Principle for Qualifiers
The fifteen elements should be usable and understandable with or without the qualifiers
Qualifiers refine meaning (but may be harder to understand)
Nouns can stand on their own without adjectives
If your software encounters an unfamiliar qualifier, look it up -- or just ignore it!
"has a“ relations break the model
E.g., a creator has a hair color
Cornell CS 502

16. Resource has Subject "Languages -- Grammar" Resource has Date
Test for “good““ qualifiers:
cover and ask:
-- Does the statement still make sense?
-- Is it still correct?
Resource
has
Subject
"Languages -- Grammar"
LCSH
Resource
has
Date
" "
ISO8601
Revised
Cornell CS 502

17. “Incorrect” Qualification
Resource
has
creator
“Cornell University”
affiliation
Resource
has
subject
“pre-schoolers”
audience
Cornell CS 502

18. Open questions in this model
Are uncontrolled and unconstrained values really useful for discovery?
Is it possible for an organization (DCMI) to control the evolution of a language?
How can "simple discovery metadata" be combined with complex descriptions? Is there a notion of graceful degradation?
Can DC serve as a lingua franca (mapping template) among more complex models
Cornell CS 502

19. Models for Deploying Metadata
Embedded in the resource
low deployment threshold
Limited flexibility, limited model
Linked to from resource
Using xlink
Is there only one source of metadata?
Independent resource referencing resource
Model of accessing the object through its surrogate
Resource doesn’t ‘have’ metadata, metadata is just one resource annotating another
Cornell CS 502

20. Syntax Alternatives: HTML
Advantages:
Simple Mechanism – META tags embedded in content
Widely deployed tools and knowledge
Disadvantages
Limited structural richness (won’t support hierarchical,tree-structured data or entity distinctions).
Cornell CS 502

21. Dublin Core in HTML
HTML constructs
<link> to establish pseudo-namespace
<meta> for metadata statements
name attribute for DC element (DC.element.ER)
content attribute for element value
scheme attribute for encoding scheme or controlled vocabulary
lang attribute for language of element value
Cornell CS 502

22. Dublin Core in HTML example
<link rel="schema.DC" href=" <meta name="DC.Title" content="Business Unusual”> <meta name=“DC.Title” lang=“es” content=“negocio inusual”> <meta name="DC.Creator" content="Carl Lagoze"> <meta name="DC.Subject" content="bibliographic control web cataloging "> <meta name="DC.Date.Created" scheme="W3CDTF"
content=" "> <meta name="DC.Format" content="text/html"> <meta name="DC.Identifier" content="
Cornell CS 502

23. Unqualified Dublin Core in XML
Cornell CS 502

24. Multi-entity nature of object description
Photographer
Camera type
Software
Computer artist
Cornell CS 502

25. Attribute/Value approaches to metadata…
The playwright of Hamlet was Shakespeare
subject
implied verb
metadata noun
literal
Hamlet has a creator Shakespeare
Playwright
metadata adjective R1
“Shakespeare”
“Hamlet”
dc:creator.playwright
dc:title
Cornell CS 502

26. …run into problems for richer descriptions…
The playwright of Hamlet was Shakespeare, who was born in Stratford
Hamlet has a creator Stratford
birthplace
“Stratford” R1
“Shakespeare”
dc:creator.playwright
dc:creator.birthplace
Cornell CS 502

27. …because of their failure to model entity distinctions
“Shakespeare”
name R1 R2
creator
birthplace
title
“Stratford”
“Hamlet”
Cornell CS 502

28. Applying a Model-Centric Approach
Formally define common entities and relationships underlying multiple metadata vocabularies
Describe them (and their inter-relationships) in a simple logical model
Provide the framework for extending these common semantics to domain and application-specific metadata vocabularies.
Cornell CS 502

29. Events are key to understanding resource complexity?
Events are implicit in most metadata formats (e.g., ‘date published’, ‘translator’)
Modeling implied events as first-class objects provides attachment points for common entities – e.g., agents, contexts (times & places), roles.
Clarifying attachment points facilitates understanding and querying “who was responsible for what when”.
Cornell CS 502

30. ABC/Harmony Event-aware metadata ontology
Recognizing inherent lifecycle aspects of description (esp. of digital content)
Modeling incorporates time (events and situations) as first-class objects
Supplies clear attachment points for agents, roles, existential properties
Resource description as a “story-telling” activity
Cornell CS 502

31. Resource-centric Metadata
Title
Anna Karenina
Author
Leo Tolstoy
Illustrator
Orest Vereisky
Translator
Margaret Wettlin
Date Created
1877
Date Translated
1978
Description
Adultery & Depression ?
Birthplace
Moscow
Birthdate
1828
Cornell CS 502

32. “Orest Vereisky”
“Leo Tolstoy”
"Moscow"
“1828”
“Margaret Wettlin”
“illustrator”
“author”
“translator”
“creation”
“1877”
“1978”
“translation”
“Russian”
“English”
“Anna Karenina”
“Tragic adultery and the search for meaningful love”
Cornell CS 502

33. Breaking the metadata bottleneck Human vs. machine generation
Simple text scraping
HTML tags as hint
Other structural methods
Natural language methods and machine learning
Contextual methods
Google (text and image search)
Cornell CS 502

34. Putting metadata in its place
Cornell CS 502

35. Query engine architecture space
Cornell CS 502