Post on 15-Jan-2015
description
NISO Webinar: Library Linked Data:
From Vision to Reality
December 11, 2013
Speakers: Jon Voss - Strategic Partnerships Director, We Are What We Do
Matt Miller - Front End Developer, NYPL Labs at the New York Public Library
Silvia Southwick - Digital Collections Metadata Librarian, UNLV University Libraries
Cory Lampert - Head, Digital Collections , UNLV University Libraries
http://www.niso.org/news/events/2013/webinars/linked_data
Linked JazzRevealing the
relationships of the jazz community
Matt Miller@thisismmillerDecember 2013
Project Overview
• Investigating the application of Linked Open Data to enhance the discovery and visibility of digital cultural heritage materials.
• Build new methods of connecting cultural data.• Uncover meaningful connections between
documents and data related to the personal and professional lives of musicians who often practice in rich and diverse social networks.
Professor Cristina Pattuelli at the Pratt Institute School of Library Information Science is the director of the project which began in 2011.
Linked Data Now!
Why?• Bootstrap your project with existing data.• Highlights knowledge you have created and
knowledge that is missing. • Facilitates sharing, but also growing your own
project.
Bootstrapping – Identifying
Research QuestionHow can we discover and analyze the rich and diverse network of
relationships between jazz musicians?
Primary SourcesOral history interview transcripts
of jazz musicians.
Bootstrapping – Identifying
Research QuestionHow can we discover and analyze the rich and diverse network of
relationships between jazz musicians?
Primary SourcesOral history interview transcripts
of jazz musicians.
We need to know the names (and variants) of jazz
musicians in a structured controlled vocabulary.
Bootstrapping – Identifying
Charlie Parker
Many different LOD datasets contain this information. We need to access, query and link it
for only jazz related individuals.
Bootstrapping – Querying
Bootstrapping – Querying
• Processing the DBpedia dataset resulted in around 9,000 URIs.– DBpedia is fluid! After each release (currently 3.9) we
reprocess the files resulting in the addition of 500-700 URIs.
• We now have a name directory, but we want additional forms of personal names. To accomplish this we try mapping to Library of Congress.
• Matching DBpedia and LC URIs is not automatic.
Bootstrapping – Mapping• We matched identities based on:
• Name• Life Dates• White listed words found in sources
(http://www.loc.gov/mads/rdf/v1#Source)
• Reconciling authorities is difficult!• Use others work: http://viaf.org/viaf/data/
• But don’t discount your own processes.• Using our relatively simple process we
were able to match about 1500 more URIs than VIAF.org.
• This is due to a smaller domain (jazz).
Our name directory creation and authority matching is documented:
https://github.com/thisismattmiller/linked-jazz-name-directory
Bootstrapping – Curating
http://linkedjazz.org/public_demo_mapping/
Bootstrapping – Review
• Start small, think big.– Specific subject domain.– Large infrastructure not required (triple stores, etc.)
• Can get started with extract files and python scripting.
• Reuse as much as possible, but try new processes leveraging domain specificity.
• Always be curating, use tools to facilitate process but a human hand is often required.
Applying the Data
• Use the name directory to locate individuals in the interview transcript.
• This project phase involves 50 transcripts.• Because the names are tied to URIs we can
infer a relationship triple between two individuals.
<foaf:Person> <rel:knowsOf> <foaf:Person>
Applying the Data
Transcript Analyzer
Transcript Analyzer
• An interface to curate the transcripts and verify detected names.
• Implements off the shelf NLP (NLTK) to attempt to locate additional names not in our directory as well as corporate names and locations.
• Global rule system, as we process more transcripts the system is being trained.
• Using URIs to represent entities we can quickly see where we are discovering new material.– 50 Transcripts
• 1800 person entities tagged.• 250 names tagged without authoritative URI.
– Knowledge Creation
New Dataset
• We have created a new LOD dataset now of jazz musician’s relationships.
• Our next steps are:– Visualize.– Further qualify the rel:knowsOf relationships.– Provide access to the data created.
Qualify Relationships – 52nd St.
• Recruit jazz experts and enthusiasts to help categorize relationships based on transcript text.
• We use existing vocabularies to build the data set: Foaf, Relationship Vocabulary, Music Ontology
• The interface is critical for crowdsourcing tools, we work with user experience experts and conduct user studies to refine our public facing tools
Qualify Relationships – 52nd St.
http://linkedjazz.org/52ndStreet/
Provide Access
• We provide a SPARQL endpoint.• But also a traditional API:
– http://linkedjazz.org/api/– Can return:
• JSON• N-Triples • Gephi graph files (GXEF)
Learn and Grow as a Team
• Experience through doing.
• Empower graduate students with skills and practical experience working with a LOD project.
• Use the project as a vehicle to make intra- and inter-intuitional collaborations.
Linked Jazz Team July 2013
Next Steps
• Refactor our prototype tools into sustainable open source projects.
• Redesign 52nd St. based on user study groups.• Work on emerging collaborations with Jazz Centers.
Linked, Exposed Data: UNLV Linked Data Project
NISO Webinar: Library Linked Data: From Vision to RealityDecember 11, 2013
Silvia B. SouthwickDigital Collections Metadata LibrarianUNLV Libraries
Cory K. LampertHead, Digital CollectionsUNLV Libraries
Agenda
• Motivation • Environment• UNLV Linked Data project• Technologies• Transforming metadata into linked data• Next steps
How it Started
• Conferences and “buzz”• Curiousity and professional development• Exploration and pilot project• Compelling results; sharing impact of what
we’ve learned• Assessment • Much more to do...
Current Practice
• Data (or metadata) encapsulated in records• Records contained in collections• Very few links are created within and/or across
collections• Links have to be manually created• Existing links do not specify the nature of the
relationships among recordsThis structure hides potential links within and across collections
What we can do with linked data
• Free data from silos• Expose relationships• Powerful, seamless, interlinking of our data• Users interact or query data in new ways• Search results would be more precise• Data can be easily repurposed
Making the Case for Linked Data in Academic Library Digital Collections
– Problem: Rich metadata is being lost in dumbed down DC records
– Issue: Investment and resource allocation (Item-level philosophy)
– Goal: Increased: exposure, collaboration, and openness
– Outcome: Increased discovery and user-focus
Gaining Buy In
Administration• Innovative project, high impact• Pilot, experiment, learn by doing, share results Staff• We already have the metadata; We need to
transform them into triples• Managing change
Graphical Representation: One Record
Examples of records
Showgirls Menus
Dreaming theSkyline
titleDecember 12, 1915
Implications (Internal)
• Cross-unit collaboration is necessary• Staff expertise will evolve• Staff roles will change to accommodate new /
parallel workflow• Data clean-up will be an investment• Management of data becomes critical• Discovery issues = user interfaces still need
development
Implications (External)
• Publish data from our collections in the Linked Data Cloud to improve discoverability and connections with other related data sets on the Web
• Sharing data in new ways with new partners may raise new issues
• Need to engage with linked data community for technologies, tools, best practices, and to demand library vendor support for LOD.
UNLV Linked Data Project
Goals: • Study the feasibility of developing a common
process that would allow the conversion of our collection records into linked data preserving their original expressivity and richness
• Publish data from our collections in the Linked Data Cloud to improve discoverability and connections with other related data sets on the Web
PROJECT IMPLEMENTATION
Actions Technologies
Prepare dataExport data
Import dataPublish
Open Refine
Mulgara /Virtuoso
CONTENTdm
Import dataClean dataReconcileGenerate triplesExport RDF
Prepare / Export Data
Technology: CONTENTdm
• Increase consistency across collections: – metadata element labels– use of CV, share local CVs– etc.
• Export data as spreadsheet
Create mapping between metadata elements and EDM model predicates
OpenRefine
• Open source
• It is a server – can communicate with other datasets via http
• Open Refine and its RDF extension should be installed
Screenshots to show some of the functions we have used
OpenRefine first screen
Facets
Split multi-value cells
Facet view forGraphic Elementsafter splitting
Reconciliation
Specifying Reconciliation service
Activating Reconciliation
Creating a Skeleton
Exporting RDF files
Actions Technologies
Prepare dataExport data
Import dataPublishQuery
Open Refine
Mulgara /Virtuoso
CONTENTdm
Import dataClean dataReconcileGenerate triplesExport RDF
Mulgara Triple Store: Import
A simple SPARQL query
Select *
where
{ ?s ?p ?o} limit 100
SPARQL: Querying Data
• Using Virtuoso PivotViewer
Query
Costume DesignDrawings
Showgirls
Next steps for the UNLV project
• Transform all digital collections into linked data (parallel structure)
• Increase linkage with other datasets• Design interfaces to access and display our data
and related data from other datasets• Evaluate alternative interfaces from user’s
perspective• Produce a cost benefit analysis to inform future
plans for the development of digital collections
Thank You!
Questions?
NISO Webinar • December 11, 2013
Questions?All questions will be posted with presenter answers on the NISO website following the webinar:
http://www.niso.org/news/events/2013/webinars/linked_data
NISO Webinar: Library Linked Data: From Vision to Reality
Thank you for joining us today. Please take a moment to fill out the brief online survey.
We look forward to hearing from you!
THANK YOU