MEOW: Metadata Enhancement and OAI Workshop

Metadata Enhancement, Sharing and OAI-PMH

What does this record describe?

identifier: http://name.university.edu/IC-FISH3IC-X0802]1004_112

publisher: Museum of Zoology, Fish Field Notes format: jpeg rights: These pages may be freely searched and displayed.

Permission must be received for subsequent distribution in print or electronically.

type: image subject: 1926-05-18; 1926; 0812; 18; Trib. to Sixteen Cr. Trib.

Pine River, Manistee R.; JAM26-460; 05; 1926/05/18; R10W; S26; S27; T21N

language: UND source: Michigan 1926 Metzelaar, 1926--1926; description: Flora and Fauna of the Great Lakes Region

Creating descriptive metadata

• “Digital library” systems– ContentDM– ExLibris Digitool– Greenstone

• Library catalogs

• Spreadsheets & databases

• XML

Some Emerging Trends in Metadata Creation

• “Schema-agnostic” metadata• Metadata that is both shareable and re-purposeable• Harvestable metadata (OAI-PMH)• “Non-exclusive”/”cross-cultural” metadata--i.e. it’s

okay to combine standards from different metadata communities--e.g. MARC and CCO, DACS and AACR, DACS and CCO, EAD and CDWA Lite, etc.

• Importance of authorities--and difficulties in “bringing along” the power of authorities with shareable metadata records

• The need for practical, economically feasible approaches to metadata creation

Building “Good digital collections”

• Interoperable – with the important goal of cross-collection searching

• Persistent – reliably accessible• Re-usable – repositories of

digital objects that can be used for multiple purposes

Issues

• Many significant digital collections lack standardized metadata

• Such inconsistent metadata causes access problems, especially when collections are aggregated

• Enhancing metadata for existing collections is often difficult to undertake because of a lack of tools to automate the process

Issues

• IMLS grant: “Enhancing and Remediating Legacy Metadata for Effective Resource Sharing.”

• “Metadata today is likely to be created by people without any metadata training … metadata records are also created by automated means … unsurprisingly, the metadata resulting form these processes varies strikingly in quality and often does not play well together .. Nevertheless, many metadata aggregators use this metadata to build services for end users, thus contributing to criticisms that metadata is of limited value, can’t be trusted or that it’s demonstrably so incomplete as to be worthless”--Diane Hillman.

Issues

• Useful services depend on good metadata, but most metadata is not very good

• Human created metadata is expensive• Automated crawling strategies are limited by:

– Accessibility barriers (rights issues, technical issues)

– Variable results with crawling technologies for non-text

• Best metadata does not rely solely on information contained within the resource itself– Ex.: Controlled vocabularies, descriptions, links

What is shareable metadata?

- Is quality metadata

- Promotes search interoperability“the ability to perform a search over diverse sets of metadata

records and obtain meaningful results.”

- Is human understandable outside of its local context

- Is useful outside of its local context

- Is machine processable

Shareable metadata defined

• Promotes search interoperability - “the ability to perform a search over diverse sets of metadata records and obtain meaningful results” (Priscilla Caplan)

• Is human understandable outside of its local context

• Is useful outside of its local context

• Preferably is machine processable

Why share metadata anyway?

• Benefits to users– Single search of a variety of digital resources– Aggregation of subject-specific resources– Higher quality resources

• Benefits to institutions– Increased user access to collection by allowing

metadata to appear in other places– Exposure to broader audience, new users– Surfacing rare, unknown, or scattered collections

Different shapes and sizes…

• Range of different aggregations:– focused subject area v. comprehensive – specialized audience v. general audience

• Range of different displays

Metadata as a view of the resource

• There is no monolithic, one-size-fits-all metadata record

• Metadata for the same thing is different depending on use and audience

• Affected by format, content, and context

• Descriptive vs. administrative vs. technical, etc. data

Metadata is a view of a resource

• No monolithic, one-size-fits-all metadata record

• The view might be different depending on use and audience as well as format, content, and context

• Content standard is a view• Metadata standard is a view• Vocabulary used is a view

Choice of metadata format(s) as a view

• Many factors affect choice of metadata formats

• Many different formats may all be appropriate for a single item

• High-quality metadata in a format not common in your community of practice is not shareable

Focus of description as a view

• Link between records for analog and digital

• Hierarchical record with all versions

• Physical with link to digital

• All versions in flat record

• Content but not carrier

Finding the right balance

• Metadata providers know the materials– Document encoding schemes and controlled

vocabularies– Document practices– Ensure record validity

• Aggregators have the processing power– Format conversion – Reconcile known vocabularies– Normalize data– Batch metadata enhancement

6 Cs and lots of Ss of shareable metadata

ContentConsistencyCoherence

ContextCommunication

Conformance

Metadata standards Vocabulary and encoding standards

Descriptive content standards Technical standards

Content

• Choose appropriate vocabularies

• Choose appropriate granularity

• Make it obvious what to display

• Make it obvious what to index

• Exclude unnecessary “filler”

• Make it clear what links point to

Consistency

• Records in a set should all reflect the same practice– Fields used– Vocabularies– Syntax encoding schemes

• Allows aggregators to apply same enhancement logic to an entire group of records

Coherence

• Metadata format chosen makes sense for materials and managing institution– Not just Dublin Core!

• Record should be self-explanatory• Values must appear in appropriate elements• Repeat fields instead of “packing” to explicitly

indicate where one value ends and another begins

Context

• Include information not used locally

• Exclude information only used locally

• Current safe assumptions– Users discover material through shared

record– User then delivered to your environment for

full context

• Context driven by intended use

Communication

• Method for creating shared records

• Vocabularies and content standards used in shared records

• Record updating practices and schedules

• Accrual practices and schedules

• Existence of analytical or supplementary materials

• Provenance of materials

Conformance to Standards

• Metadata standards (and not just DC)

• Vocabulary and encoding standards

• Descriptive content standards (AACR2, CCO, DACS)

• Technical standards (XML, Character encoding, etc)

Before you share…

• Check your metadata– Appropriate view?– Consistent?– Context provided?– Does the aggregator have what they need?– Documented?

Can a stranger tell you what the record describes?

The reality of sharing metadata

• We can no longer afford to only think about our local users

• Creating shareable metadata will require more work on your part

• Creating shareable metadata will require our vendors to support (more) standards

• Creating shareable metadata is no longer an option, it’s a requirement

So where does RDA fit in?

• RDA is a content standard

• MODS is a metadata standard

• RDA is closely aligned with MARC and MODS

• Useful to have a RDA – MODS examples particularly as MODS is shifting away from MARC

• Although because of its origins in the library world RDA presupposes MARC as a vehicle for cataloging records, this emergent cataloging code could also be used with MODS, Dublin Core, or other metadata schemas.

• Lack of engagement with the MODS community beyond the Library of Congress

DLF Aquifer Guidelines

• Does NOT recommend any one content standard over another

“Choice and format of titles should be governed by a content standard such as the Anglo-American Cataloging Rules, 2nd edition (AACR2), Cataloguing Cultural Objects (CCO), or Describing Archives: A Content Standard (DACS). Details such as capitalization, choosing among the forms of titles presented on an item, and use of abbreviations should be determined based on the rules in a content standard. One standard should be chosen and used consistently for all records in an OAI set.”

Metadata aggregators

• CIC Metadata Portalhttp://cicharvest.grainger.uiuc.edu/

– Records and digital resources shared by consortium of institutions, provided for

– Educators, researchers, and general public

Benefits:

Single comprehensive search of multiple collections and a variety of disciplines

Metadata aggregators

• National Science Digital Libraryhttp://nsdl.org

- Online resources and records pertaining to science & math education and research, vetted for inclusion, provided for

- Educators, researchers, policy makers, and the general public

Benefits:Single portal serving a range of resources on a

specialized topic to a diverse audience

OAI-PMH

• Open Archives Initiative-Protocol for Metadata Harvesting

• OAI-PMH defines a mechanism for harvesting records containing metadata from repositories.

• The OAI-PMH gives a simple technical option for data providers to make their metadata available to services, based on the open standards HTTP (Hypertext Transport Protocol) and XML (Extensible Markup Language).

• The metadata that is harvested may be in any format that is agreed by a community (or by any discrete set of data and service providers), although unqualified Dublin Core is specified to provide a basic level of interoperability

OAI-PMH

• Metadata from many sources can be gathered together in one database, and services can be provided based on this centrally harvested, or "aggregated" data.

• Data Provider: a Data Provider maintains one or more repositories (web servers) that support the OAI-PMH as a means of exposing metadata.

• Service Provider: a Service Provider issues OAI-PMH requests to data providers and uses the metadata as a basis for building value-added services. A Service Provider in this manner is "harvesting" the metadata exposed by Data Providers

OAI-PMH Structure

• Intentionally designed to be simple• Data providers

– Have metadata they want to share– “Expose” their metadata to be harvested

• Service providers– Harvest metadata from data providers– Provide searching of harvested metadata from

multiple sources– Can also provide other value-added services

Data Providers

• Set up a server that responds to harvesting requests

• Required to expose metadata in simple Dublin Core (DC) format

• Can also expose metadata in any other format expressible with an XML schema

Service Providers

• Harvest and store metadata• Generally provide search/browse access

to this metadata• Can be general or domain-specific• Can choose to collect metadata in formats

other than DC• Generally link out to holding institutions for

access to digital content• OAIster is a good example

Pop Quiz

What is the OAI-PMH?

(Select one answer)

(a) The OAI-PMH is a protocol for sharing metadata.

(b) The OAI-PMH is a low-barrier protocol for searching across repositories and retrieving resources from them.

Multiple Service Providers can harvest from

multiple Data Providers.

Basic functioning of OAI-PMH

Finding the right balance

• Metadata providers know the materials– Document encoding schemes and controlled

vocabularies– Document practices– Ensure record validity

• Aggregators have the processing power– Format conversion – Reconcile known vocabularies– Normalize data– Batch metadata enhancement

Why share metadata?

• Benefits to users– One-stop searching– Aggregation of subject-specific resources

• Benefits to institutions– Increased exposure for collections– Broader user base– Bringing together of distributed collections

Don’t expect users will know about your collection and remember to visit it.

Why share metadata with OAI?

• “Low barrier” protocol• Shares metadata only, not content,

simplifying rights issues• Same effort on your part to share with one

or a hundred service providers (basically)• Wide adoption in the cultural heritage

sector• Quickly eclipsed older methods such as

Z39.50

Common Problems with Metadata in Aggregation

• Consistency

• Sufficiency

• Compatibility

Consistency problems

• Appearance of data• Application of format• Granularity of records• Vocabulary usage

Result:Service Provider must normalize data

(if can determine what “normal” is)

Sufficiency problems

• Too little info for understanding what resource is, especially outside of local context

Result:Users don’t know whether a resource is

relevant or not

Compatibility problems

Information in records is– Erroneous– Unnecessary– Incompatible

Result:

Interferes with harvesting and indexing

Common content mistakes

• No indication of vocabulary used – Names

• LCNAF: Michelangelo Buonarroti, 1475-1564• ULAN: Buonarroti, Michelangelo

– Places• LCSH: Jakarta (Indonesia) • TGN: Jakarta

– Subjects• LCSH: Neo-impressionism (Art)• AAT: Pointillism

• Shared record for a single page in a book• Link goes to search interface rather than item being described• “Unknown” or “N/A” in metadata record

Common context mistakes

• Leaving out information that applies to an entire collection (“On a horse”)

• Location information lacking parent institution

• Geographic information lacking higher-level jurisdiction

• Inclusion of administrative metadata

Different collections describing same source:

1. Duplication problem

Duplication problems

• Duplicates – can be identical records, – can describe the same source, but with

different metadata, and– can describe the same source but with links to

different or slightly different location identifiers (e.g. index page vs. splash page).

2. Record quality could be different

No date.

No format, type, grade level, language, rights, etc.

Description might be based on the then table of contents.

No date.

Description might be based on the then cover page. Description is no longer relevant.

Format missed images.

No grade level, rights.

No date.

Description is more general and can last longer. +

Recorded grade level, format, type +

No language, rights.

3. Incomplete data causes low recall

Completeness Search & Display

No FORMAT information

•Collections which did not provide FORMAT information are excluded from being searched at advanced searches

Completeness Search & Display (cont.)

• Collections did not use EDUCATIONLEVEL or AUDIENCE element would not be in the pool when a user searches by Grade Level.

Completeness Search & Display (cont.)

Expectations for discovery systems are rising

• Growth of cutting-edge systems outside of libraries affecting user expectations

• Higher user expectations are a good thing!

• Many expected functions will be easier with robust structured metadata– Genre access– Faceted browsing– Limiting scope by time, place, etc.

Libraries are having trouble meeting those expectations

• Non-textual resources are more difficult to search

• Legacy metadata isn’t always structured in ways that allow high-level services

• Legacy metadata doesn’t always include enough information to allow high-level services

• Creating new metadata needed to provide high-level services is prohibitively expensive

Enter automated enhancement methods

• Much research has been done• Little of it has been put into production systems

in library metadata creation environments– Still requires human intervention– Fear of human skills becoming devalued

• Metadata aggregators, out of necessity, are among the first implementers

• Automatic enhancement holds great promise for standardizing and streamlining metadata creation and aggregation activities

Why Enhance Metadata at All?

• Four categories of problems limit metadata usefulness:– Missing data: elements not present– Incorrect data: values not conforming to

proper usage– Confusing data: embedded html tags,

improper separation of multiple elements, etc.– Insufficient data: no indication of controlled

vocabularies, formats, etc.

record

record

Metadata Repository

Enriched record

Aggregation

Solving the problems -- Enriching and enhancing

harvested records

Aggregation

It is possible that these problems be eliminated to certain level through a process called ‘aggregation’ in a metadata repository.

The notion behind this process is that a metadata record, “a series of statements about resources,” can be aggregated to build a more complete profile of a resource.

NSDL Metadata Repository

iViaEnhancement

Service

Provider Aorig metadata<dc:title><dc:identifier><dc:creator><dc:type>

ENCEnhancement

Service

<dct:audience source=ENC><dct:educationLevel source=ENC>

NSDL normalized/augmented<dc:title source=A><dc:creator source=A>

<dc:subject GEM source=iVia><dc:subject LCSH source=iVia><dc:subject LCC source=iVia>

<dc:identifier URI source=MR><dc:type DCMIType source=MR>

Provider A

iVia enhancements<dc:subject GEM><dc:subject LCSH><dc:subject LCC>

OAI

Safe xform enhancements<dc:identifier URI><dc:type DCMIType>

ENC enhancements<dct:audience><dct:educationLevel>

OAI

NSDL SafeTransforms

OAI

OAI

NSDL Metadata Repository

iViaEnhancement

Service

Provider Aorig metadata<dc:title><dc:identifier><dc:creator><dc:type>

ENCEnhancement

Service

<dct:audience source=ENC><dct:educationLevel source=ENC>

NSDL normalized/augmented<dc:title source=A><dc:creator source=A>

<dc:subject GEM source=iVia><dc:subject LCSH source=iVia><dc:subject LCC source=iVia>

<dc:identifier URI source=MR><dc:type DCMIType source=MR>

<dct:audience source=ENC><dct:educationLevel source=ENC>

NSDL normalized/augmented<dc:title source=A><dc:creator source=A>

<dc:subject GEM source=iVia><dc:subject LCSH source=iVia><dc:subject LCC source=iVia>

<dc:identifier URI source=MR><dc:type DCMIType source=MR>

Provider A

iVia enhancements<dc:subject GEM><dc:subject LCSH><dc:subject LCC>

OAI

Safe xform enhancements<dc:identifier URI><dc:type DCMIType>

ENC enhancements<dct:audience><dct:educationLevel>

OAI

NSDL SafeTransforms

OAI

OAI

missed?! missed?!

Incorrect element mappingOPTIONS mapped to SUBJECT, missing all KEYWORDS

Data-conversion-related problems

missing keywords

AUTHOR mapped to DESCRIPTION, no CREATOR

Only physical description, no content description

Inappropriate mapping

CLASSIFICATION mapped to SUBJECT and missed all the KEYWORDs.

Solving the problems -- Correcting the errors

• Checking and testing crosswalks!!!

• Re-harvesting

• Training how to use OAI tools

Value space that should follow standardized rules: Examples from values associated with DATE element

• 1979• 2000-03• 2000-03-01• 2001-01-

02T21:48.00Z• 200003• C1999, 2000• January, 1919• May, 1919

• 1987, c2000• ?1999• 1952 (issued)• (1982)• 1930?]• 1823-1845• Between 1680 and

1896?dc.date: Recommended best practice for encoding the date value is defined in a profile of ISO 8601 [W3CDTF] and includes (among others) dates of the form YYYY-MM-DD

Value space that should apply standard controlled vocabularies: Examples from values associated with

LANGUAGE element

• en• eng• en-GB• en-US• English

• engfre• new• Korean• Deutsch• German

LOCLANGUAGE:: German

dc.language: Recommended best practice is to use RFC 3066 [RFC3066] which, in conjunction with ISO639 [ISO639]), defines two- and three-letter primary language tags with optional subtags.

Two efforts to promote shareable metadata

• Best Practices for OAI Data Provider Implementations and Shareable Metadata– http://webservices.itcs.umich.edu/mediawiki/oaibp/?T

ableOfContents

• Digital Library Federation / Aquifer Implementation Guidelines for Shareable MODS Records – http://www.diglib.org/aquifer/dlfmodsimplementationg

uidelines_finalnov2006.pdf

Everyone could use better metadata!

?

Metadata Enhancement

• Clustering and classification

• Automated name authority control

• Date normalization

• Thumbnail generation and creating actionable URLs

Clustering and classification

• UC-Irvine and Michigan

• Evaluate topic/subject-based metadata enhancement

• Clustering: “learning the topics” (pre-process)

• Classification: using the learned topics to determine topics in records and records in topics

Topic Model

• State-of-the-art statistical algorithm

• Learns a set of topics or subjects covered by a collection of text records

• Works by finding patterns of co-occurring words

• Determines the mix of topics associated with each record

Clustering and classification

• Mix of scientific repositories• Average of 75 words per record• Used words from <title>, <description>,

<subject> for clustering• Only kept words that occurred in more

than 10 records• Result: a final vocabulary of 90,000 words• Cluster words into topics: ended up with

500 topics

Clustering and classification

• 500 topics too many to look at

• Needed to organize topics under broad topical categories– Cluster the clusters (automatic)– Use pre-defined categories

• Classify group of keywords (manual + automatic)• Create hierarchy by hand (manual)

Cluster

Classify

OAIrecords

vocab-ulary

preprocesstopic

model(cluster/learn)

topics

clustering is learning the

topics

OAIrecords

vocab-ulary

preprocesstopic

model(classify)

1. topics in records2. records in topics

classification is using the

learned topics

Preprocessing Example• <ID=oai:CiteSeerPSU:44072>

• <title>Reinforcement Learning: A Survey

• <description>This paper surveys the field of reinforcement learning from a computer-science perspective. It is written to be accessible to researchers familiar with machine learning. Both the historical basis of the field and a broad selection of current work are summarized. Reinforcement learning is the problem faced by an agent that learns behavior through trial-and-error interactions with a dynamic environment. The work described here has a resemblance to work in psychology, but differs considerably in the details and in the use of the word "reinforcement." …

• <subject>Leslie Pack Kaelbling, Michael Littman, Andrew Moore. Reinforcement Learning: A Survey

<ID=oai:CiteSeerPSU:44072>

reinforcement learning survey

survey field reinforcement learning computer science perspective written accessible researcher familiar machine learning historical basis field broad selection current summarized reinforcement learning faced agent learn behavior trial error interaction dynamic environment resemblance psychology differ considerably detail word reinforcement …

leslie pack kaelbling littman andrew moore reinforcement learning survey

vocab-ulary

preprocess

Example Topics (1)

Words in Topic Topic Label

gene sequence genes sequences cdna region amino_acid clones encoding cloned coding dna genomic cloning clone

gene sequencing

social cultural political culture conflict identity society economic context gender contemporary politic world examines tradition sociology institution ethic discourse

cultural identity

general_relativity gravity gravitational solution black_hole tensor einstein horizon spacetime equation field metric vacuum scalar matter energy relativity

relativity

house garden houses dwelling housing homes terrace estate home building architecture residence homestead residences road cottage domestic fences lawn historic

domestic architecture

Example Topics (2)

Words in Topic Usefulness

large small size larger smaller sizes scale sized largest

Reasonable but unusable

foi para pacientes por foram dos doen resultados grupo das tratamento entre

Topic about patient treatment, in Spanish

building street visible santa_ana view avenue public_library front orange corner

Not usable: mix of concept words and specific geographic location words

Topics Assigned to a Record

Metadata Record Topic Labels

(% words assigned)

Aggregating sets of judgments: two impossibility results compared. (C. List and P. Pettit)

May's celebrated theorem (1952) shows that, if a group of individuals wants to make a choice between two alternatives (say x and y), then majority voting is the unique decision procedure satisfying a set of attractive minimal conditions ...

game theory (21%)

argument (12%)

criteria (7%)

Clustering and classification

• Selected useful topics– [ t482 ]     labor worker employment wage

market labour job unemployment wages earning panel find evidence individual participation skill

– [ t372 ]   firm investment capital productivity innovation industry sector economic industrial foreign industries corporate ownership technological companies evidence

Clustering and classification

• Selected less useful topics– [ t255 ]     journal author chapter vol notes

editor publication issue special bibliography reader references appendix literature submitted topic

– [ t013 ]   university department mail edu institute science california email usa computer york fax college press center address

Broad Topical Categories (BTCs)

• By clustering the clusters– Worked well– Can choose desired number of BTCs

• By classifying groups of keywords– Worked well too

• Then review and manually edit– Include or exclude any subtopic

Clustering and classification: Further evaluation

• Need to test non-English and cultural heritage repositories

• Need usability testing

• “On the horse” problem more prevalent

• When to re-cluster?

Automated Name Authority Control (ANAC)

• Johns Hopkins University: research only; never implemented

• 29,000 Levy sheet music records

• 13,764 unique names

ANAC• The evidence used to determine the probability of a match between a name to an LC

record is a set of Boolean tests involving the name, the Levy metadata associated with that name, and the LC record.

• The following fields were used by ANAC: • Levy record:

– Given name: often abbreviated – Middle names: often abbreviated – Family name – Modifiers: titles and suffixes – Date: publication year – Location: publication location (city)

• LC record: – Given name: includes abbreviations – Middle names: includes abbreviations – Family name – Modifiers: titles and suffixes – Birth: year of birth – Death: year of death – Context: miscellaneous data

ANAC

• The tests used are: first name equality and consistency, middle name equality and consistency, music terms present in LC record context, name modifier consistency, Levy sheet music publication consistent with LC author birth and death, and Levy record publication location in LC record context

ANAC• In order to train the system, the Cataloging Department at the

Sheridan Libraries generated ground truth data.• For each name in 2,000 randomly selected Levy metadata records,

catalogers recorded the authorized form of the name when a matching authority record was available.

• The entire process required 311 hours (approximately seven minutes per name).

• The human catalogers used much the same type of evidence as ANAC in establishing matches. Catalogers examined name similarity; compared publication dates from the Levy records to birth and death dates in the authority records; and examined authority record note fields for musical terms.

• In addition, the catalogers often searched for bibliographic records of other editions of a particular title to determine the authoritative name assigned to the subject.

ANAC

• Overall, ANAC was successful 58% of the time. When a name had an LC record, ANAC was successful 77% of the time, but when an LC record did not exist for a name ANAC was successful only 12% of them time. The reason for this discrepancy is that ANAC cannot learn whether or not a name has been added to the LC authority file.

• It took ANAC five hours and forty-five minutes to classify the 2,673 (2,841 minus 168) names, or about eight seconds per name. The database-bound process of retrieving the candidate set of MARC records given a family name consumed most of this time.

ANAC

• Matching very dependent on contextual data• Machine matching much faster than manual (8 sec. vs. 7

min.)• Performance reasonable even with dirty metadata. • Machine matching could enhance manual work• Combination of machine processing and human

intervention produced best results• Approach could be tweaked by comparing names to

multiple authority files or domain specific databases• ANAC not a generalizable tool, but there are others

Date Normalization

• How to make “ca. 1880” a machine-readable date but not a 19-2 baseball score

• California Digital Library• Created for American West project, so

sidestepped issue of B.C.E. date normalization• Uses <date> element• If no <date>, searches for date-like strings in

<title>, then <description>• Currently normalizes to YYYY only, not MM or

DD (will add later)

<date>: Encoding Variances

• 1991-10-01 • ca. 1920. • (ca). 1920) • 2001.06.08 by CAD • Unknown • ca. June 19, 1901. • (ca). June 19, 1901) • [2001 or 2002.] • 1853. • c1875. • c1908 November 19

• c1905 • 1929 June 6 • [between 1904 and 1908] • [ca. 1967] • 1918 ? • [1919 ?] • 191-? • 1870 December, c1871 • 1920, 1921, 1922, 1923,

1924, 1925, 1926, 1927, 1928, 1929

Normalization Process

• Extract dates

• Standardize approximate dates, e.g. ca. 1902 = 1902~– CDL uses +/- 5 years e.g. 1902~ = 1897-1907

• Normalize Dates

• Populate date.found or date.guess

• Create era, decade and year tokens

Recognizing Unknowns

• Recognizes wide range of expression of date unknownness in <date> element, e.g. unknown, unkn, unavail, n.d., nd, undated, no date, not indicated

• Looks for date-like strings to normalize in <title>, then <description> when <date> element contains one of these expressions

• If no date, look for Civil War, Renaissance, dates of reigns of sovereigns, etc.

Known Issues

• Distinguishing “c” for circa from “c” for copyright (currently interprets as later)

• Getting tripped up by baseball scores in non-<date> elements

• Getting tripped up by 4-digit item identifiers

Thumbnail generation and creating actionable URLs

• Thumbnail: “A miniature representation of a page or image that is used to identify a file by its content”—PC Magazine

• Need coordination between metadata harvesting and thumbnail grabbing

• Digital libraries need digital objects

• <identifier> element in DC is a problem: hard to identify link to actual object

Thumbnail generation and creating actionable URLs

• “Users should be able to download, manipulate, morph, annotate, cross-search, and repurpose digital library content”

• Find best possible link, find best possible image, build thumbnail: registry of links; data providers retain the responsibility of maintaining the authoritative version of their resource

• Need to find a way to express intellectual property rights related to manipulation of objects

• Try to get providers to supply better metadata, but in mean time use what we’ve got

Characteristics of quality metadata:

• Completeness. -- choosing an element set allowing the resources in question to be described as completely as is economically feasible, and -- applying that element set as completely as possible.

• Accuracy. -- the metadata being correct and factual, and conforming to syntax of the element set in use.

• Provenance. Here provenance refers to the provision of information about the expertise of the person(s) creating the original metadata, and its transformation history.

• Conformance to expectations. Metadata elements, use of controlled vocabularies, and robustness should match the expectations of a particular community.

• Logical consistency and coherence. -- element usage matching standard definitions, and consistent application of these elements.

• Timeliness. Currency--metadata keeping up with changes to the resource it describes. Lag -- a resource’s availability preceding the availability of its metadata.

• Accessibility. Proper association of metadata with the resource it describes and readability by target users contribute to this characteristic.

Additional characteristics that make quality metadata more useful in a shared environment:

• Proper context. … each record contain the context necessary for understanding the resource the record describes, without relying on outside information.

• Content coherence. … need to contain enough information such that the record makes sense standing on its own, yet exclude information that only makes sense in a local environment.

• Use of standard vocabularies. The use of standard vocabularies enables the better integration of metadata records from one source with records from other sources.

• Consistency. All decisions made about application of elements, syntax of metadata values, and usage of controlled vocabularies, should be consistent within an identifiable set of metadata records so those using this metadata can apply any necessary transformation steps without having to process inconsistencies within such a set.

• Technical conformance. Metadata should conform to the specified XML schemas and should be properly encoded.

How Can We Ensure A Better Quality?

• Make Policies on:– minimum quality requirements, – quality measurement instruments, – quality enforcement policies, – quality enhancement actions, and – the training of metadata creators.

• Training! – A 2-hour training may eliminate hundreds of errors

• IT team should talk with content team– A test of crosswalk for OAI harvest may prevent thousands of mis-matched

or missed values• Use Tools:

– Provide instructions on best practices– Use template for inputting records, with suggested syntax, vocabularies, and

build-in values– Use validators– Implement duplicate checking algorithm

Final Thoughts

• Creating shareable metadata requires thinking outside of our local box

• Creating shareable metadata will require more work on our part

• Creating shareable metadata will require our vendors to support (more) standards

• Creating shareable metadata is no longer an option, it’s a requirement

Before we share…

• Check our metadata– Appropriate view?– Consistent?– Context provided?– Does the aggregator have what they need?– Documented?

Can a stranger tell you what the record describes?

More thoughts

• Automated metadata enhancement techniques promise to play an essential role in building and aggregating digital library collections

• But they are not a “magic bullet” – must be used together with other techniques– User-contributed metadata– Content-based retrieval– Item-level attention by specialists

• Many collections could benefit– Legacy collections described in MARC– Special collections largely undescribed, especially at the item

level• Should technical services expand metadata activities• Catalogers and their skills essential to this process

The Way Forward?

• Service providers should be more demanding (i.e. require that data providers adhere to certain standards and use certain vocabularies, require “pre-washed” metadata.

• Data providers should consistently use appropriate standard schemas in their local systems.

• Service providers should consider “adding value” via services like vocabulary mapping, query expansion, vocabulary-assisted searching, user-added metadata, post-harvest subsetting, metadata enhancement, etc

Lessons Learned

• Metadata (descriptive, technical, rights, administrative, preservation) is one of your biggest investments.

• Do it once, do it right (consistent schemas, controlled vocabularies), and you can re-purpose metadata in a wide variety of ways.

• Good descriptive metadata records can be core—records don’t need to be “full” to be “good.”

• Creation of consistent, standards-based descriptive metadata (a.k.a. cataloging!) is time- and labor-intensive, but it’s worth it.