Smart collections

Smart collections
Pru Mitchell & Richard Leibbrandt
pru.mitchell@esa.edu.au
leib0006@csem.flinders.edu.au

The goal
To describe resources for education so learners and educators can find
the right resource at the right time

every reader his/her book
Ranganathan 1931, 5 principles of library science

Educational value
Curriculum relevance
Intended user
Educational level
Educational quality
How can I use this for learning?

Research question
Can artificial intelligence tools and techniques assist with discovering, evaluating and tagging digital learning resources?

Research partners
Flinders University Artificial Intelligence and Language Lab
V.A.L.I.A.N.T.
Education Network Australia (edna)

Proof of concept
3 people
$30,000
3 months
3 show and tells
blog for peer review

Research team
Dr Richard Leibbrandt
Dr Dongqiang Yang
Darius Pfitzner
Prof David Powers
Pru Mitchell
Sarah Hayman
Helen Eddy

Technical concepts
artificial intelligence
text classification (TC)
semantic annotation
Text Frequency/Inverse Document Frequency (TFIDF)
semantic web
taxonomy and folksonomy

DSpace metadata tool
Text
Extraction tool
Subject Keywords
Resource
edna Categories
html
edna-userlevel
edna-audience
Educational use elements

Key phrase extraction
a (the full text document)
?
b (keywords, audience, userlevel)
Which terms in the text prompt this decision?

Findings
labour-intensive challenge
subject classification easier than userlevel
audience and user level indicated by
meaning, vocabulary, choice of words, style, font size, graphic elements, layout
‘reading’ web pages is a complex literacy

Semantic network
Automated subject analysis
Can system predict edna category?
Tools
WordNetWikipediaDbpedia

Findings

User terms to thesaurus
Schools Online Thesaurus
scot.curriculum.edu.au
Can system predict ScOT terms from teacher keywords?

User metadata
the only group that can categorize everything is everybody
Clay Shirky, 2005Ontology is overrated

Mapping teacher tags
Proper nouns & brands
.NET, iPod, excel
Semantic ambiguity
notes -> Pitch (Music)
Stemming
practice exams -> Practical Examinations

Findings
% ScOT terms predicted from keywords

Benefits
Efficiency of cataloguing through classification suggestions
Improved user experience through more relevant and consistent search results
Improved integration of user contributed resources if tags are mapped to taxonomy

Conclusions
Artificial intelligence systems showed some success in subject categorisation of text-based digital learning resources
Key phase extraction to support subject categorisation was more successful than of audience and user-level categorisation

Summary
Automated classification based on artificial intelligence may be useful as a means of supplementing and assisting human classification, but is not at this stage a replacement for human classification

Future
ScOT development
Achievement Standards Network
Machine Readable Curricula
Multilingual thesauri
OER ‘travelling well’ globally

Credits
Foyer artwork 2008, State Library of South Australia http://www.slsa.sa.gov.au/site/page.cfm?u=409
Climate change timeline, Copyright ABC
http://www.abc.net.au/environment/timeline.html
Holeymoon 2008, Rent 1, 2, 3 CC-by-nc-sahttp://www.flickr.com/photos/holeymoon/2926989641
International Standard Book Number, Wikipedia, CC-by-sahttp://en.wikipedia.org/wiki/International_Standard_Book_Number
Leave your mark, Oxfam Australia
http://www.leaveyourmark.my3things.org
O’Connor, D 2005 Binary Finary, CC-by-nc-sahttp://www.flickr.com/photos/clockwerx/9267076
Ranganathan, S 1931, The five laws of library science The Madras Library Association, London: Goldston, Madras
Yelkrokoyade 2010, Conservera de Lisboa , CC-sahttp://commons.wikimedia.org/wiki/File:Conservera_de_Lisboa.jpg

Smart collections

by Pru Mitchell on Oct 14, 2010

Accessibility

Categories

Tags

Upload Details

Usage Rights

Statistics

Smart collections — Presentation Transcript