1 of 38

New Models Require

New Action Plans

Implementing Linked Data Within PCC

Steven Folsom, Cornell

<http://orcid.org/0000-0003-3427-5769>

PCC PoCo Meeting

November 1st, 2017

Presentation Link: http://bit.ly/2xJU7Nb

2 of 38

Prompt for Infrastructure White Paper

“...draft a white paper on what the emerging, post-MARC infrastructure might be like… a high-level view of the range of possibilities for that environment. Not a prediction or a prescription of how it ought to be arranged, but an overview of some of the possibilities and the necessary entities and relationships…”

3 of 38

Acknowledgements

Special thanks to:

Linked Data Advisory
Chew Chiat Naun (Harvard, $c Mentor and guide)
Michelle Durocher (Harvard, $c Mentor and guide)
Sarah Ross (Cornell, $c Principal Cataloger and Truth-teller)
LD4L (the wolves that raised me in the wilderness)

4 of 38

What is Linked Data?

5 of 38

What from our existing workflows do we want to port forward?

6 of 38

Centralized vs. Distributed Models

Paul Baran’s 1964 publication, On Distributed Communications: Introduction to Distributed Communication Networks, (prepared for the United States Air Force Project RAND) describes network models in terms of their vulnerabilities. Baran’s now famous diagram characterizing centralized, decentralized, and distributed models is useful to illustrate the spectrum of options library communities have to communicate about our collections and share data.

Linked data allows us to explore models closer to distributed end of the spectrum, which as Baran points out, is the least vulnerable to large scale outages, but it also allows us to this ability to explore having the agency of saying things locally and still have the data available globally for others to use.

Centralized allows for:

Gatekeeping to allow for quality assurance
Obvious where large scale batch enhancements occur

Decentralized

No single point of failure
Allows for flexibility, choice in services
More agency at the local level to assert information, but on a global stage.

7 of 38

Challenges our infrastructure will need to address

8 of 38

discovery (lookups) and caching (copies) of data

9 of 38

Diverse APIs and Data Flows

With sharing MARC records we’re usually talking about Z39.50, MARC Record Loads through FTP, maybe OAI PMH.

Linked data publishers have many options on how they serve their linked data.

Dereferenceable URIs (classic definition of linked data).
We often hear about Triplestores and Sparql Endpoints
There is Linked Data Fragments
RDF dumps

�I won’t go into detail about the various options, but because each has strengths and weaknesses.

We often find data providers offering one or more ways of accessing the data. Which means consumers of the data have to accommodate the various APIs in order to ingest the data.

As pointed out in the Task Group on Identity Management, there are members of the semantic web community attempting to define standard protocols like WebSub to make it easier to communicate when and how datasets have been updated.

10 of 38

Diversity of Bibliographic Data Models

11 of 38

Application Profiles and Quality

Define the floor, not ceiling
“Roomy” enough for all needs
wrt RDA, not all data will conform

Lurking underneath some of these decisions about bibliographic models is a tension between content standards and the ability to interoperate with web data.

We’re in a moment now, where there’s a loud discussion about the value of RDA and metadata practices in general.

“To ensure a smooth transition to linked data, we need to show sensitivity to our cataloging colleagues, who have invested heavily in RDA training; ultimately, there will be a transition.

This evolution is not beyond the ability or experience of catalogers; large portions of library cataloging practice is already outside the bounds of strict RDA, especially if we consider our non-MARC workflows.

Although RDA has been widely adopted in the MARC community,the constraints of the MARC format mean that current RDA implementations are in complete in their scope and highly colored by legacy practices.

At the same time, RDA itself is based on a model that would arguably benefit from reevaluation in the light of semantic web best practices.”

-

https://wiki.duraspace.org/display/LD4P/bibliotek-o?preview=/79795231/90964891/bibliotek-o_and_RDA201708.pdf

12 of 38

Linked Data Validation: SHACL as a partial answer

ex:LiteraryWarrantShape a sh:NodeShape ;

sh:scopeClass foaf:Person ;

sh:property [

sh:path foaf:name ;

sh:minCount 1;

sh:maxCount 1;

sh:datatype xsd:string ;

] ;

sh:property [

sh:path ex:createdWork ;

sh:minCount 1;

sh:maxCount 1;

sh:nodeKind sh:IRI ;

] .

:TressieMcMillanCottom a foaf:Person ;

foaf:name “Tressie McMillan Cottom” ;

ex:createdWork :LowerEd .

:Russell a foaf:Person ;

foaf:name “Maria Doria Russell” ;

ex:createdWork “The Sparrow” .

ex:JoshRitter a foaf:Person ;

foaf:firstName “Josh” ;

ex:createdWork :TheAnimalYears .

There machine actionable ways to define our profiles that can support validation.

With linked data there are different types of validation that become important.

URI validation: Is the URI still working? Does it supply linked data?

Validating quality and completeness of the data:

Without a record to describe, we need new techniques and tools to measure and mark the quality of data in the graph. In MARC, we say that this record conforms to a certain standard/encoding level or this set of records conforms to a certain standard.

With linked data we can describe at a dataset level levels of quality.

There are ways to assert the shape requirements for individual items in a dataset using things like SHACL (Shapes Constraint Language) or ShEx (Shape Expressions).

There are tests we can add to our tools in order to test whether we are getting the desired input and output.

13 of 38

Reconciliation

<http://orcid.org/0000-0003-3427-5769> ,

owl:sameAs

<http://vivo.cornell.edu/individual/sf433> ,

<http://viaf.org/viaf/316560733> ,

<http://worldcat.org/entity/person/id/2630057950> .

<http://id.loc.gov/authorities/names/no2015079947>

mads:identifiesRWO

<http://orcid.org/0000-0003-3427-5769> .

14 of 38

“Round-tripping”

or... how we’ll need MARC records for the foreseeable future.

Even if everything was in place for a library to switch over to linked data today, for the foreseeable future we will need supply MARC records for various cooperative cataloging agreements we have, and data aggregator requirements for MARC.

That, combine with the fact that our ILSs continue to require MARC for ordering/receiving/bill paying functions, means that we’ll need to be able to create operational MARC records to perform these functions.

The focus should be on “operational” because round-tripping with lossiness isn’t possible, so a minimum core record will be all we can support in this round trip conversion.

One of the goals of the PCC URIs in MARC Task Group is to help make these conversions more meaningful and less lossy.

There is a lot of work still to do in terms of URI best practices in MARC.

More work to define best practices in the Authority Record
Consideration of new maintenance models both for the backfile and specifically authorities.

We’ve been waiting for Phase 3b a while.
Certain authority files get updated and are available more quickly than others.

15 of 38

Hypothesis: Possible Models and Services

16 of 38

decentralized

publishing

PCC

LD4L

LD4P

IMLS Shared Authorities

OCLC

Ex Libris

Casalini

FOLIO

Hydra

VIVO

Getty

ISNI

id.loc.gov

Orcid

17 of 38

decentralized data discovery as a service

18 of 38

Data flow

PCC

LD4L

LD4P

IMLS Shared Authorities

OCLC

Ex Libris

Casalini

FOLIO

Hydra

VIVO

Getty

ISNI

id.loc.gov

Orcid

19 of 38

What does it mean for

the cataloger?

Not that different

from what we

have now.

20 of 38

Under the hood

we linked to external

resources AND possibly

publish in our own

namespace.

21 of 38

Distributed Vocabulary Maintenance

As a means to be more responsive to current needs and social responsibility.

I’m acutely aware of where I’m standing, and who my hosts are. I’m grateful for everything the Library of Congress is able to accomplish in service of the community.

I also aware that how we conceive of vocabulary and entity maintenance is changing.

There’s a greater emphasis on faceted vocabularies.

There’s also a desire to be able to create entities at the point of need.

A part of the community trying to consider what decentralized models for vocabulary maintenance might look like, and if a term or hierarchy within a vocabulary doesn’t represent a concept the way a local institution or group desires, how can they mint their own terms without breaking the whole ecosystem.

We see this the discussions around the LCSH headings “Illegal Aliens”, “Climatic Changes”, but this comes up in less controversial scenarios as well.
There’s a need for both the re-thinking maintenance models of centralized vocabularies and also enabling decentralized “local” terms (perhaps aligned with community vocabularies) that can be shared more broadly for like-minded institutions.

22 of 38

“Recordless environment”

Manage at the Triplestore level

Named Graphs (like “a record”)

Linked Data Notifications (LDN)

23 of 38

How do we get there?

24 of 38

Safe Assumptions

Assume decentralized publication of RDF
Assume (at least for the time being) heterogeneous data models
Decentralized discovery of RDF will be provided.
Support expandable/contractible scope/discovery of the data we link to
Understand the many existing APIs for data sources
Reconciliation will be needed
Profiles and Validation will become important in production environments

25 of 38

The Role of the PCC

26 of 38

Next steps

Evaluate the data models we expect to use, provide feedback

Evaluate the data we would expect to link to

Understand the related models so that we can better specify how systems should interact with external data

27 of 38

Next steps

With community and model defined, define shared application profiles.

The models aren’t enough.
Tune our application profiles through experience.

28 of 38

Next steps

As we consider our desired data models and data sources understand who our partners then become, and what those partnership models should look like.

Ideal data models are open, transparent.
Ideal data sources are open, transparent.
Ideal partnership models are open, transparent.
Who’s in our circle of trust and why?

29 of 38

Next steps

Partner with Vendors and Open Source Community to gain experience with early tools (don’t wait for perfection)

Define and lobby for functional requirements
This includes all metadata platforms, not just linked data platforms

30 of 38

Next steps

Consider how the PCC committee structure and working groups are constructed and communication channels.

31 of 38

The Working Groups

ISNI Pilot

BIBFRAME Task Group

Task Group on URIs in MARC

Task Group on the Work Entity

PCC Steering Committee

Standing Committee on Standards

PCC Policy Committee

Standing Committee on Applications

Standing Committee on

Training

Linked Data Advisory Group

Operations Committees

Task Group on Identity Management in NACO

32 of 38

Organizing the work

33 of 38

Staffing. Who pays for it?

Should such important work be staffed almost entirely through volunteerism?

How do we train and recruit talent so that we’re equipped to progress conversations the conversations we’re having and solve problems?

What is the connection between the work of the PCC and how PCC Member Libraries hire and assign staff?

34 of 38

Training: pet ALL the puppies

35 of 38

Iterative Workplans

36 of 38

Respectfully and with gratitude.

Steven Folsom (Cornell, $c not a guru)

<http://orcid.org/0000-0003-3427-5769>

37 of 38

Image Credits

S:5 https://www.loc.gov/catworkshop/courses/naco-RDA/Module%201-NACO%20Foundations.pptx

S:6 https://www.rand.org/content/dam/rand/pubs/research_memoranda/2006/RM3420.pdf

S:8 http://www.popastro.com/help/help.php?title_pag=starting%20to%20use%20your%20telescope

S:9 http://www.JoeTourist.net/, via Wikimedia Commons

38 of 38

Image Credits

S:10 https://twitter.com/azaroth42/status/856560201314344962

S:11 https://www.dwell.com/article/10-best-modern-showers-to-inspire-your-bathroom-renovation-f8dac0c2

S:25 https://paulgordonbrown.com/2017/06/27/100-questions-you-can-use-for-intentional-conversations-in-the-residence-halls/

S:34 http://beariscool.blogspot.com/2015/10/

S:35 http://chiphouston.com/five-important-questions-love/