Enhancing social tagging with a knowledge organization system

1
Enhancing social tagging with a knowledge
organization system

• Brian Matthews
• STFC

2
Outline

• Who are STFC ?
• Controlled Vocabulary
• Social Tagging
• EnTag
• Aims
• Glamorgan/UKOLN/Intute Experiment
• STFC Experiment
• SKOS

3
Science and Technology Facilities Council

• Provide large-scale scientific facilities for UK
  Science
• particularly in physics and astronomy
• E-Science Centre at RAL and DL
• Provides advanced IT development and services to
  the STFC Science Programme
• Also includes library and institutional
  repository
• Strong interest in Digital Curation of our
  science data
• Keep the results alive and available
• RD Programme
• DCC, CASPAR
• EnTag

4
Controlled Vocabulary

• Traditional way of providing subject
  classification
• For shelf-marking
• For searching
• For association of resources
• Several different types used, such as
• Subject Classification
• Keyword lists
• Thesaurus
• Each has different characteristics

5
HASSET (I)

• UK Data Archive, Univ of Essex
• Humanities and Social Science Electronic
  Thesaurus
• Some 1000s of terms
• Structure based on British Standard
  57231987/ISO 2788-1986 (Establishment and
  development of monolingual thesauri).
• preferred terms, broader-narrower relations,
  associated terms
• http//www.data-archive.ac.uk/search/hassetSearch.
  asp

6
HASSET (II)
7
HASSET (III)
8
Observations on using controlled vocabularies

• Precise classification of resources
• Good for precision and recall
• Can exploit the hierarchy to modify query
• Using the broader/narrower/related terms
• Highly expensive
• Requires investment in specialist expertise to
  devise the vocabulary
• Requires investment in specialist expertise to
  classify resources.
• Hard to maintain currency

9
Social Tagging

• The Web 2.0 way of providing search terms
• People tag resources with free-text terms of
  their own choosing
• Tags used to associate resources together
• del.icio.us, flickr
• Folksonomy
• the terms a community choses to use to tag its
  resources.

10
Connotea
11
Connotea sharing tags
12
Connotea Tag Cloud
13
Observations on Social Tagging

• People often use the same tags or keywords (e.g.
  Preservation, Digital Library)
• this makes things which mean the same thing to
  people easier to find
• Cheap way of getting a very large number of
  resources marked up and classified
• Represents the community consensus in some
  sense
• The Wisdom Of Crowds
• Has currency as people update
• Tag clouds of popular tags
• However, people often use similar but not the
  same tags
• e.g. Semantic Web, SemanticWeb, SemWeb, SWeb
• People make mistakes in tags
• mispellings, using spaces incorrectly.
• Some tags are more specific than others
• E.g. controlled vocabulary, thesaurus, HASSET
• People often associate the same words together
  with particular ideas in images
• these are captured in clusters

14
EnTag Project

• Enhanced tagging for discovery
• JISC funded project
• Partners
• UKOLN
• University of Glamorgan
• STFC
• Intute
• Non-funded
• OCLC Office of Research, USA
• Danish Royal School of Library and Information
  Science
• Period 1 Sep 2007 -- 30 Sep 2008
• http//www.ukoln.ac.uk/projects/enhanced-tagging/

15
EnTag Background

• Controlled vocabularies
• Improve information retrieval and discovery
• But, costly to index with, especially the amount
  of digital documents
• Require subject and classification experts
• Social tagging
• Holds the promise of reducing indexing costs
• Uses terms describing how people see the resource
• Serendipity
• But, tags uncontrolled,
• missed associations
• Relating different views
• Highly personal (me, important),
• Quality and ranking
• Depth of term

16
EnTag Purpose

• Investigate the combination of controlled and
  social tagging approaches to support resource
  discovery in repositories and digital collections
• Aim to investigate
• whether use of an established controlled
  vocabulary can help move social tagging beyond
  personal bookmarking to aid resource discovery

17
EnTag Objectives

• Investigate indexing aspects when using only
  social tagging versus when using social tagging
  in combination with a controlled vocabulary
• In particular, does this lead to
• Improve tagging
• Relevance of tags (perspective, aspects,
  specificity, exhaustivity, terminology
  (linguistic level, semantic level, contextual
  level)
• Consistency
• Efficiency (time used, user satisfaction)
• Use (tags selected, clouds consulted, order of
  consultation)
• Improve retrieval
• Effectiveness (degree of match between user and
  system terminology)
• In two different contexts
• Tagging by readers
• Tagging by authors

18
Testing Approach

• Main focus
• free tagging with no instructions
• Versus
• tagging using a combined system and guidance for
  users
• Two demonstrators
• Intute digital collection http//www.intute.ac.uk
• Major development
• Tagging by reader
• DDC
• STFC repository http//epubs.cclrc.ac.uk/
• Complementary development
• Tagging by author
• A more qualitative approach

19
Intute
20
Intute demonstrator searching
21
Intute demonstrator basic tagging
22
Intute demonstrator enhanced tagging
23
EnTag Intute user study (II)

• Test setting
• 50 graduate students in political science
• 60 documents, covering up to four topics of
  relevance for the students
• Data collection
• Logging time spent, selection patterns,
• Pre- and post-questionnaires

24
EnTag Intute user study (I)

• Test comparison of basic and advanced system
• Indexing
• Perspective, specificity, exhaustivity
• Linguistics (word class, single word/compound,
  spelling, language)
• Consistency
• Efficiency (time used, user satisfaction)
• Use (tags selected, clouds consulted, order of
  consultation)
• Retrieval efficiency
• Degree of match between user and system
  terminology
• user tags, DDC tags, controlled Intute keywords,
  title terms, text terms

25
STFC Case Study EPubs
26
STFC demonstrator
27
STFC Author study

• A study on a Authors of papers
• Smaller number - c.10-12.
• Regular depositors ( gt 10 papers each)
• Subject experts
• Expect that they would want their papers
  accurately tagged so that they are precisely
  found
• A more qualitative study

28
Expected Feedback

• Relative value of tagging vs. controlled terms
• Does it give more satisfactory (accurate,
  consistent) tags?
• Does it lead to the consideration of tags they
  would not have thought of?
• Do they select deeply in the hierarchy?
• Is this something they would like to see
  supported more, and would use?
• Is it worth the overhead?
• How we should use a combination of tagging and
  controlled vocab in our system ?
• To Be Continued..

29
Building a Web of Knowledge

• Social tagging and controlled vocabulary
  complement each other
• Tagging entry level, quick, does the job, but
  error prone, fuzzy
• Controlled vocabulary, accurate, but slow and
  expensive
• Use one to leverage the other
• Use both to build a Web of knowledge
• The things in the world and their link via their
  subjects
• Get the users to build the means of organising
  the knowledge

30
http//purl.org/net/aliman
30
31
SKOS Simple conceptual relationships
32
Conclusions

• Controlled vocabulary and Tags complement each
  other
• Hope to get some interesting evidence over the
  next month as the studies are complete.
• Web 2.0 world offers the possibility of
  combining these results
• SKOS a format to use both tags and controlled
  vocabulary as part of the Web of Linked Data
• Also use Web 2.0 to build the vocab themselves.

33

• Questions?
• b.m.matthews_at_rl.ac.uk

34
EnTag Enhanced tagging for discovery

• Research collaboration between Glamorgan
  University, UKOLN, INTUTE, CCLRC, OCLC, and DB
• Financed by JISC Capital Programme
• Research goal
• Investigation of the combination and comparison
  of controlled and folksonomy approaches to
  semantic interoperability supporting resource
  discovery in repositories and digital collections
  
• Evaluation in two communities of use at Intute
  (Social science), focussing on tagging by readers
  (postgraduate users), and at CCLRC, focussing on
  tagging by authors
• The two studies are carried out as separate
  projects
• Intute project use DDC as controlled vocabulary
• Evaluation by quantitative and qualitative
  measures

35
Evaluation Intute focus and objective

• Context tagging as part of information
  searching and relevance assessment, tagging for
  recommendation and sharing
• Hybrid system investigate whether tagging can
  be improved by a combination of traditional tag
  clouds and clouds of controlled descriptors,
  including interactive tools such as tag
  suggestions, access to browsing of DDC, etc.
• Improve tagging
• Relevance of tags (perspective, aspects,
  specificity, exhaustivity, terminology
  (linguistic level, semantic level, contextual
  level)
• Consistency
• Efficiency (time used, user satisfaction)
• Use (tags selected, clouds consulted, order of
  consultation)
• Improve retrieval
• Effectiveness (degree of match between user and
  system terminology)