1. The Technical Infrastructure of the NSDL Dean Krafft, Cornell University dean@cs.cornell.edu

2. NSDL Technical Overview Structure of the talk:  NSDL 1.0 Architecture and Lessons Learned  The Fedora-based NSDL Data Repository (NDR) and NSDL 2.0  Inspiring Contribution and Collaboration - ExpertVoices  Other NSDL 2.0 Services and Tools  Q&A

3. What is the NSDL?  An NSF-funded $20 million/year program in Science, Technology, Engineering and Mathematics (STEM) education  A digital library describing over a million carefully selected online STEM resources from over 100 collections (at http://nsdl.org)  A core integration team (Cornell, UCAR, Columbia) working with 9 “pathways” portals and over 200 NSF grantees  A large community of researchers, librarians, content providers, developers, students, and teachers

5. What are the building blocks?  A distributed set of NSDL collections  A central repository of aggregated information about Science, Technology, Engineering, and Mathematics digital resources  A set of services that build on the repository, initially search and archive  A set of portals that expose NSDL STEM resources to a variety of user communities

6. Infrastructure overview: NSDL 1.0 STEM Collections on the Web Central Metadata Repository Search Service Archive Service Collection Registration System NSDL.org Portal Protocol: OAI-PMH HTTP REST SQL

7. NSDL 1.0 Ingest  Create a “union catalog” of Dublin Core metadata records for STEM resources  Harvest those records from collections using OAI-PMH (openarchives.org)  Normalize and augment metadata records (change to qualified DC)  Store records in an Oracle DB and re- serve qualified DC through OAI-PMH

8. Metadata Aggregation … Collection OAI Servers Metadata Repository (Oracle DB) OAI-PMH Server Metadata Ingest Service Harvest Management/ Collection Registration System OAI-PMH NSDL Dublin Core NSDL DC

9. NSDL 1.0 Search  Harvest metadata records from MR using OAI-PMH  Crawl URLs specified in the metadata using Nutch  Build a search index using metadata plus full-text of available content pages  Expose search in a web portal at nsdl.org for K-gray access to NSDL resources

10. Search Architecture Web Content Pathway Portal Pathway Portal NSDL.org Portal Lucene Query Engine REST Query Interface OAI Harvester OAI-PMH Search and Discovery Server Nutch Harvester http/ftp Lucene index generator Metadata Repository …

11. NSDL 1.0 Lessons  Rather than one portal for everyone, support communities with common interests: Pathways now provide discipline and area-specific portals  Metadata is expensive: unlike traditional libraries, e.g. through OCLC, digital collections have very “mixed quality” metadata, with unusual and inconsistent coding  On the good side: Oracle DB and OAI-PMH server scaled successfully to over 1 million catalog records

12. NSDL 1.0 Lessons continued  OAI-provided collections need 3 types of expertise: domain (resources & pedagogy), metadata (vocabulary & formatting), and technical (XML schema, UTF8, HTTP, OAI-PMH).  In many cases it took several months from first contact to successful OAI harvest, and the average harvest failure rate has stayed at 25%-50%, with only 23% of that transient failures  Incremental harvesting fundamental to efficient processing, but problematic: issues with persisting deleted records and recovering from partial harvests  Result: some automation, but high people cost

13. NSDL 1.0 Summary  Metadata Repository was quick to implement using known technologies, but  Limited model  Metadata-centric orientation  No content – only metadata  Limited relationships – collection/item  Limits on context, structure, and access  Severe limits on contribution and collaboration  One-way data flow: NSDL → Users

14. Going beyond the card catalog  Create an NSDL that guides not just resource discovery, but resource selection, use, and contribution  Supports creating “context” for resources  Presents resources in context: in a lesson plan; with ratings; correlated with education standards  Supports creating a permanent archive of resources  Enables community tools for structuring, evaluation, annotation, contribution, collaboration  Goal: Create a dynamic, living library

15. NSDL 2.0: NSDL Data Repository  Goals:  Architecture of participation: service-based, not a monolithic application/single user experience  Remixable data sources and data transformations  Harnessing (and capturing) collective intelligence  A free market of millions of inter-related resources (create the “long tail”)  Two-way data flow: NSDL ↔ users  Solution: Fedora-based NSDL Data Repository

16. Fedora: the NDR middleware  A Flexible, Extensible Digital Object Repository Architecture (http://www.fedora.info)  Open source project with $2.2 million in Mellon funding 2002-2007  Collaboration of Cornell and Univ. of Virginia  Key funded users include:  eSciDoc project (collaboration of the Max Planck Society and FIZ Karlsruhe)  VTLS Corp., Harris Corp., Library of Congress  Australian Research Repositories Online to the World (ARROW)  Royal Library Denmark, National Library, and DTU

17. The Fedora Vision: A Repository for Rich Information Networks

18. What is Fedora?  An architecture, toolkit, and implementation: middleware, not a vertical application  DSpace in contrast: a vertical application with a fixed workflow targeted at users  Stores arbitrary internal and external digital objects, disseminations (transformations and combinations), relationships among objects  Entirely SOAP/REST based, disseminations are URLs  XML data store; RDBMS cache; RDF triplestore supports relationship queries

19. Fedora Key Features  Content aggregation  Digital object model to combine information entities in novel ways  Knowledge integration  Ontology-based relationships among objects  Information reuse  Create secondary, tertiary objects  Information transformation  Combine objects with computational services  Collaboration and contribution  Enable annotation, info sharing, workflow, contextualization  Information management and preservation  XML-based object storage  Service-oriented architecture; web services  Store relationships and service linkages with objects

20. Digital object identifier Reserved Datastreams Key object metadata Disseminators Web-service methods for distributing views of recombined content Datastreams Set of content or metadata items (local or external URL redirects) Fedora Digital Object Model Component View Persistent ID (PID) Dublin Core (DC) Datastream Audit Trail (AUDIT) Relations (RELS-EXT) Disseminator Default Disseminator

21. Implementing the NDR with Fedora  Multiple Object Types:  Resources (with local or remote content)  Metadata  Aggregations (collections)  Metadata Providers (branding)  Agents  Relationships with arbitrary graph queries:  Structural (part of)  Equivalence  Annotation

23. Draft NDR API Characteristics  Uses REST calls for all interactions  Specializes Fedora for NDR objects/relationships  Disseminations allow combining metadata from multiple sources, or related content  Authentication: Requests signed with private key associated with an agent  Authorization: Agent can become a metadata provider or aggregator; can create resources  Documentation being developed at http://ndr.comm.nsdl.org

24. NDR Architecture

25. An Information Network Overlay  Think of the NDR as a lens for viewing science content on the net  Content can be:  Local: stored directly in the NDR  Remote: accessed through a URL  Computed: derived from a database or web service  Archived: an older version stored at SDSC  It all has a repository-based URL

26. Network Overlay View User View API/UI Repository View with Relations & Annotations Resources on the Web

27. Status of the NDR  Repository in test load  over 875,000 metadata records  over 2 million digital objects  Over 163 million RDF triples (lots)  Scaling challenges  moved to 64-bit architecture with 32GB memory  need to carefully structure RDF queries  can scale current system by factor of five  need to move to more powerful triplestore (Oracle)  Estimating fully operational beta version of new NDR in June

28. How should we use the NDR?  The NDR provides powerful capabilities for:  Creating context around resources  Enabling the NSDL community to directly contribute resources and context  Representing a web of relationships among science resources and information about those resources  How do we use it? Here’s one specific example …

29. Issues in STEM Education  Issue: Need to support scientific inquiry  Issue: Students need a better understanding of the processes of scientific research  Issue: Teachers are often under- prepared to teach science and math  Issue: Scientists need tools to make science and math research more available

30. Addressing the Needs  In Response: NSDL is building an educational tool that…  Models scientific inquiry and exposes the processes of scientific research  Promotes and facilitates conversations between research and education communities  Brings content expertise into the classroom to support under-prepared teachers  Allows scientists, teachers, and media specialists to collaboratively develop instructional context around NSDL resources

31. ExpertVoices

32. What is Expert Voices?  A system using blogging technology to:  Support STEM conversations among scientists, teachers and students  Tie NSDL resources to real-world science news  Create context for resources to enhance discovery, selection and use  Enable NSDL community members to become NSDL contributors: of resources, questions, reviews, annotations, and metadata  Expert Voices ≠ LiveJournal  Contributors are carefully selected, contributions are about science, the process of science, and education

33. Expert Voices As An Educational Tool  Topic-based discussion (e.g. tsunamis) with pointers to related resources  Research outreach (Criterion 2) – explaining and documenting NSF-funded research  Experts can add resources with topical context to the NSDL  Resources can be reviewed and annotated  Question/answer and discussion forum: scientist ↔ teacher ↔ student ↔ librarian

34. Broadening Participation: An Expert Voices Learning Scenario  “Hurricane Season Blog” run by a National Weather Service hurricane expert, an Earth Science teacher, and a school media specialist familiar with NSDL resources  Expert creates an entry for Hurricane Gertrude  “On track to hit Ft. Lauderdale in 72 hours”  “Currently undergoing eyewall replacement cycle”  “Expecting 15 foot storm surge”  Media specialist adds links to NSDL resources: Hurricane Hunters site, latest satellite photos, and USGS flooding and flood plain site (storm surge context)  Teacher makes connections to relevant standards and appropriate pedagogy for use by other teachers  Students experience engaging real-time, real-world applications of science lessons

35. Broadening Participation: An Expert Voices Outreach Scenario  NSF grantee: Bioluminescence researcher wants to make research K-12 accessible  Creates an Expert Voices conversation  Enables his students and researchers to document process and results – how science really works  Writes about publications and educational resources (e.g. www.photobiology.info)www.photobiology.info  Adds these to the NSDL, creating audience-level metadata  Entries serve as annotations that create K-12 context for the college-level research

36. Expert Voices Implementation  Open source multi-user blogging system  Published entries become NSDL resources  Owner controls publication of entries and visibility of comments  Entries can contain linked references to NSDL resources, references to URLs that should become resources, and new resource metadata  Integrated with NSDL community sign-on

37. Expert Voices Implementation  Initial blog system is multi-user WordPress  WordPress plug-ins provide NDR integration and Shibboleth authentication  Publication of blog entry creates:  Content, as a new resource with simple metadata  New NDR resources  New metadata for any referenced resources in content  Graph of relationships between entry and all referenced resources  Blog available as independent RSS feed

38. NDR Entry for Expert Voices Blog Entry New Metadata New Audience MD Referenced New Resource 1 Referenced Existing Resource 2 Annotates Metadata for Member of Metadata Provider Metadata Provider Existing Collection Topic- based Blog Member of Inferred relationship between resources

40. But Expert Voices is just the beginning…

41. NDR Application: OnRamp  NDR-integrated multi-user, multi-project content management system  Supports NSDL single sign-on and group management  Decentralized workflow for the creation and distribution of both simple and complex content  Disseminates content in multiple publication and online forms  Delivery estimated 3Q06

43. NDR Application: Instructional Architect  Created by Mimi Recker and colleagues at Utah State University  Teacher develops a lesson plan, incorporating NSDL and other resources  Assigns subject, grade level, ed standard  Distributes to class or public  Available now, with NDR integration in design

45. NDR Application: Integrated Wiki  Community of approved contributors (e.g. teachers, librarians, scientists) are granted edit access on OpenNSDL wiki  New resources and metadata are created as wiki pages and reflected into the NDR  Non-wiki-based NDR resources and metadata are displayed as read-only wiki pages, subject to comment and linking  User and project pages organize NDR resources

47. NDR Application: Content Assignment Tool  Developed by Anne Diekema, Elizabeth Liddy, et al. at the Syracuse University Center for Natural Language Processing  Uses text analysis and machine learning to suggest Educational Standards alignment for resources  Content expert assigns standard, and system learns from the assignment  Standalone tool available now; standards associated with resources in the NDR by 3Q06

48. Content Assignment Tool

49. Other applications in development  Automated grade-level assignment based on vocabulary analysis (San Diego Supercomputer Center)  iVia-based Expert-Guided crawl: Tool for Pathways and others to turn websites into resource collections (UC Riverside)  Automated subject assignment (UC Riverside)  MyNSDL: Bookmark and tag STEM resources within and outside the NSDL (Cornell)

50. … NSDL 2.0 Ecosystem Protocol: OAI-PMH HTTP REST NDR API STEM Collections Search Service Archive Service Fedora- based NDR

51. What does this mean for the user?  NSDL 2.0 applications situate resources in context, aiding both discovery and use  Users become contributors, adding new resources, ratings, annotations, and organizational structure – frequently as a side effect of using the library  Specialized portals, tagging, and powerful relationship queries and filtering support user- specific “views” into the library

52. Summary  NSDL 1.0 created a large, production digital library of STEM resources for education.  NSDL 2.0 and its tools allow scientists, mathematicians, teachers, engineers, librarians, and students to create a unique web of context, contribution, and collaboration around the high-quality STEM education resources at the core of the NSDL.

53. Acknowledgements  NSDL NSF Program Officers  Lee Zia  David McArthur  NSDL Core Integration Team  UCAR: Kaye Howe, PI and Executive Director  Cornell: Dean Krafft, PI  Columbia: Kate Wittenberg, PI  Fedora Development Team  Cornell: Sandy Payette & Carl Lagoze  Univ. of Virginia: Thornton Staples

54. Questions?

55. Contact Information Dean B. Krafft Cornell Information Science 301 College Ave. Ithaca, NY 14850 USA dean@cs.cornell.edu This work is licensed under the Creative Commons Attribution-NoDerivs 2.5 License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nd/2.5/ or send a letter to Creative Commons, 543 Howard Street, 5th Floor, San Francisco, California, 94105, USA.http://creativecommons.org/licenses/by-nd/2.5/