Ontological Foundations for Scholarly Debate Mapping Technology

1
Ontological Foundations for Scholarly Debate
Mapping Technology
Neil BENN, Simon BUCKINGHAM SHUM, John DOMINGUE,
Clara MANCINI
COMMA 08, 29 May 2008
2
Outline

• Background Access vs. Analysis
• Research Objectives
• Debate Mapping ontology
• Example Representing analysing the Abortion
  Debate
• Concluding Remarks

3
Access vs. Analysis

• Need to move beyond accessing academic documents
• search engines, digital libraries, e-journals,
  e-prints, etc.
• Need support for analysing knowledge domains to
  determine (e.g.)
• Who are the experts?
• What are the canonical papers?
• What is the leading edge?

4
Two KDA Approaches

• Bibliometrics approach
• Focus on citation relation
• Thus, low representation costs (automatic
  citation mining)
• Network-based reasoning for identifying
  structures and trends in knowledge domains (e.g.
  research fronts)
• Tool examples CiteSeer, Citebase, CiteSpace

5
CiteSpace
6
Two KDA Approaches

• Semantics
• Multiple concept and relation types
• Concepts and relations specified in an ontology
• Ontology-based representation to support more
  intelligent information retrieval
• Tool examples ESKIMO, CS AKTIVE SPACE,
  ClaiMaker, Bibster

7
Bibster
8
Research Objectives

• None considers the macro-discourse of knowledge
  domains
• Discourse analysis should be a priority other
  forms of analysis are partial indices of
  discourse structure
• What is the structure of the ongoing dialogue?
  What are the controversial issues? What are the
  main bodies of opinion?
• Aim to support the mapping and analysis of debate
  in knowledge domains

9
Debate Mapping Ontology

• Based on logic of debate theorised in Yoshimi
  (2004) and demonstrated by Robert Horn
• Issues, Claims and Arguments
• supports and disputes as main inter-argument
  relations
• Similar to IBIS structure
• Concerned with macro-argument structure
• What are the properties of a given debate?

10
Ex Using Wikipedia Source
11
Issues
12
Propositions and Arguments
13
Publications and Persons
14
Explore New Functionality

• Features of the debate not easily obtained from
  raw source material
• E.g. Detecting clusters of viewpoints in the
  debate
• A macro-argumentation feature
• As appendix to supplement (not replace) source
  material
• Reuse citation network clustering technique

15
Reuse Mismatch

• Network-based techniques require single-link-type
  network representations
• Similarity assumed between nodes
• Typically co-citation as similarity measure

16
Inference Rules
Co-authorship
Co-membership

• Implement ontology axioms for inferring other
  meaningful similarity connections
• Rules-of-thumb (heuristics) not laws

17
Inference Rules
Mutual Dispute
Mutual Support

• All inferences interpreted as Rhetorical
  Similarity in debate context
• Need to investigate cases where heuristics
  breakdown

18
Applying the Rules
19
Cluster Analysis
Visualisation and clustering performed using
NetDraw
20
Debate Viewpoint Clusters
21
Reinstating Semantic Types
BASIC-ANTI-ABORTION-ARGUMENT
BASIC-PRO-ABORTION-ARGUMENT
ABORTION-BREAST-CANCER-HYPOTHESIS
BODILY-RIGHTS-ARGUMENT
DON_MARQUIS
JUDITH_THOMSON
ERIC_OLSON
PETER_SINGER
EQUALITY-OBJECTION-ARGUMENT
CONTRACEPTION-OBJECTION-ARGUMENT
DEAN_STRETTON
RESPONSIBILITY-OBJECTION-ARGUMENT
MICHAEL_TOOLEY
TACIT-CONSENT-OBJECTION-ARGUMENT
Visualisation and clustering performed using
NetDraw
22
Two Viewpoint Clusters
BASIC-PRO-ABORTION-ARGUMENT
JUDITH_THOMSON
PETER_SINGER
DEAN_STRETTON
JEFF_MCMAHAN
JEFF_MCMAHAN
ERIC_OLSON
DON_MARQUIS
BASIC-ANTI-ABORTION-ARGUMENT
23
Concluding Remarks

• Need for technology to support knowledge domain
  analysis
• Focussed specifically on the task of analysing
  debates within knowledge domains
• Ontology-based representation of debate
• Aim to capture macro-argument structure
• With goal of exploring new types of analytical
  results
• e.g. clusters of viewpoints in the debate (which
  is enabled by reusing citation network-based
  techniques)

24
Limitations Future Work

• The ontology-based representation process is
  expensive (time and labour)
• Are there enough incentives to makes humans
  participate in this labour-intensive task?
• Need technical architecture (right tools,
  training, etc.) for scaling up
• Viewpoint clustering validation
• Currently only intuitively valid
• Possibility of validating against positions
  identified by domain experts
• Matching against philosophical camps identified
  on Horn debate maps of AI domain

25

• Thank you