Recommendations in iLumina by myfa9xT

VIEWS: 1 PAGES: 10

									Developing Recommendation Services
        for a Digital Library
    Recommendations in iLumina
 with Uncertain and Changing Data
                  Gary Geisler (UNC), Dave McArthur and
                         Sarah Giersch (Eduprise)




 Funding provided by the National Science Foundation
  DLI-Phase 2
 NSF Award #0002935
 A Digital Library of Reusable Science and Math Resources
 for Undergraduate Education
                    iLumina’s Vision
The situation:
• e-learning will create a huge demand for high-quality digital
  course content
• Publishers will meet some of this demand through ebooks
• Instructors will still need informal digital modules to complement
  and tailor ebooks to specific courses
The opportunity:
• The Internet now provides a scalable way to implement peer-
  centric sharing of informal digital content
• iLumina is a digital library that will realize this opportunity
           iLumina’s Main Themes
• Featuring diverse, small-scale SMETE resources,
  especially ones created by instructors
• Using IMS’s rich and standard metadata to describe
  resources
• Developing services underpinned by IMS metadata
  (e.g., recommendations as well as search and
  browse)
• Promoting reuse and repackaging of resources by
  faculty authors
• Implementing a partially centralized (metadata) and
  partly distributed (content) architecture
                        Architecture Overview
        Existing Repositories                      Informal Collections             Individual Contributions

                         …
                                                                       …
     Addison CSTC SECDL        Other                   UNCW Mathwright Other             iLumina Open   Physlets
     Wesley

                 *
                                                          Cataloging Tool
                     Mapping
                      Tool
                                                        Content
                                                        Review
                           *

                                                       Cataloging Tool
                                                               final
     Basic Search
   Faceted Search                           iLumina
       Browsing                             metadata
   Flexible Retrieval             Search
                                 Services                Construction       Author Cataloging
                                                                           Resource Integrator
Suggestion Boxes                                           Services
 General forums                             User
User ratings/review       Community
Formal peer review         Services
Recommendations
 Personalizations
               Information Sources for
                  Recommendations
  Data Type                Description              Quality    Availability
Resource          IMS metadata elements and         High       Complete
Characteristics   values
Resource          Minimal acceptability judgments   Variable   Partial
Evaluation        Informal user ratings
                  Formal peer reviews
Resource          User downloads                    Low        Partial
Usage             User resubmissions
User Profile      Demographics                      Variable   Partial
                  Affiliations
                  Areas of interest
                  Service preferences
            Recommendation Schemes
 Scheme Name               Sources                          Rule
Simple Winner         Evaluation (User      If a resource is highly rated,
                      ratings or reviews)   recommend it.
Profile Match         Characteristics &     If a resource matches areas of
                      Profile               interests, recommend it
Basic Content-        Self ratings &        If a resource is structurally similar
based                 Characteristics       to ones previously liked,
                                            recommend it.
Basic Collaborative   Self ratings & User   If a resource is liked by those with
Filtering             ratings               similar tastes, recommend it
  Recommendation Schemes (continued)
 Scheme Name              Sources                         Rule
Weak Content-        Self usage &         If a resource is structurally similar
based                Characteristics      to ones previously used,
                                          recommend it.
Weak Collaborative   Self usage & Other   If a resource is used by those with
Filtering            usage                similar downloads, recommend it
Content-based        Self ratings &       If a resource is similar to ones
Collaborative        Characteristics &    previously liked and liked by those
Filtering            User ratings         with similar tastes, recommend it.

Home Run             Self ratings &       If a resource is similar to ones
                     Characteristics &    previously liked and liked by those
                     User ratings &       with similar tastes, and matches
                     Profile              areas of interest, recommend it.
Challenges to Recommendation Schemes

• How can multiple recommendation schemes be
  implemented?
• Do more specific schemes work better than less
  specific ones?
• How should the choice of schemes be conditioned by
  data source quality and quantity (costs and benefits)?
• How should multiple purposes be factored into
  recommendations?
      Towards Solutions to Challenges
• Do natural experiments:
   – Evaluation, Use and Profile information will grow over time in
     iLumina
   – Test the efficacy of different schemas as data becomes richer
   – Devise schemes that effectively incorporate new information
     sources
• Consider extending profiles to include multiple
  purposes and goals
   – Index user ratings and usage data by purpose
              References and Sources
• Our project website: http://www.ilumina-project.org
• Project papers on the site:
   – JCDL: Developing Recommendation Services for a Digital Library
     with Uncertain and Changing Data. Joint Conference on Digital
     Libraries (JCDL)
   – CACM: Towards a Sharable Digital Library of Reusable Teaching
     Resources: Roles for Rich Metadata Communications of the ACM,
     Special Digital Libraries Issue, April 2001
   – JERIC: A Sharable Digital Library of Reusable Teaching
     Resources: Roles for Rich Metadata. Journal on Education
     Resources in Computing (JERIC).
   – JLibAdmin: Library Services Today and Tomorrow:
     Lessons from iLumina, a Digital Library for Creating and Sharing
     Teaching Resources. Journal of Library Administration.

								
To top