Conceptualisation of Communication in Computation: A Case Study in Formal Concept Analysis

1
Conceptualisation of Communication in
Computation A Case Study in Formal Concept
Analysis

• Peter Eklund1

1 School of Economics and Information Systems The
University of Wollongong Northfields Avenue, NSW
2522 peklund_at_uow.edu.au
2
Talk Overview

• What is a concept in formal concept analysis?
• Does formal concept analysis support C.S.
  Peirces pragmatism?
• How is formal concept analysis consistent with
  the theory of the communicative action?
• Show me how it works!

3
Basics of FCA 4
Start with Earth B small, near, moon(s) A
Earth, Mars (A,B) is a formal concept of
(G,M,I) iff
TABLE 1
4
Basics of FCA 4
Earth, Mars,small, near, moons
Start with Earth B small, near, moon(s) A
Earth, Mars (A,B) is a formal concept of
(G,M,I) iff
5
Basics of FCA 4
6
Other Common Scales
Nominal
Ordinal
Interordinal
Boolean
7
Biordinal
8
(No Transcript)
9
Extensions of FCA 8,10
Prediger Wille 8 Dau, 2003 10
10
Conceptual Knowledge Processing
Eklund and Wille (2004)
How did we come to this?
11
Pragmatic Philosophy

• According to Peirce, meaning is a triadic
  relation between a sign, an object, and an
  interpretant. 10
• According to pragmatic philosophy and C.S.
  Peirce, knowledge is formed in an unbounded
  process of human thinking, arguing and
  communicating in this connection, reflection on
  the effects of thought is significant and real
  experiences stimulate to re-thinking time and
  again. In this process, form and content are
  related so closely that they may not be separated
  without loss. Wille 2006

Charles Sanders Peirce (1839-1914)
10 http//www.angelfire.com/md2/timewarp/peirce.
html
12
Theory of communicative action 1
Jürgen Habermas

• C.S. Peirces pragmatism is continued in modern
  philosophy by Apel 2 and Habermas 1
• Habermas Communicative Action Theory helps
  explain conceptual knowledge processing and why
  it is so unlike AI.
• TCA focuses on coordinated action through
  understanding, not manipulation and control
• Action is oriented towards understanding

• This is called Communicative Rationality
• Communicative rationality is different from
  instrumental means-end rationality
• We argue that CKP supports inter-subjective
  argumentation

1 J. Habermas, The Theory of Communicative
Action, Vol. 1 Reason and the Rationalization
of Society, Beacon Press, Boston, 1984. 2
K.-O.Apel Das Apriori der Kommunikationsgemeinsch
aft und die Grundlagen der Ethik. In
Transformation der Philosophie. Band 2 Das
Apriori der Kommunikationsgemeinschaft. Suhrkamp
Taschenbuch Wissenschaft 165, Frankfurt 1976.
13
Re-thinking AI
...it is more obvious not to give up reasoning
entirely, but rather to break with the concept of
reasoning which is orientated by the pattern of
logic-mathematical proofs.. Apel 5, p.
19. Only the process of discourse and
understanding in the intersubjective community
leads to comprehensive states of rationality.
Such a process does not exclude
logic-mathematical proofs, but they can be only
part of a broader argumentative discourse. Wille
6
5 K.-O.Apel Begründung. In H.Seiffert,
G.Radninzky (Hrsg.)Handlexi-kon der
Wissenschaftstheorie. Ehrenwirth, Müunchen 1989,
14-19.
6 R.Wille Conceptual landscapes of knowledge
a pragmatic paradigm for knowledge processing.
In G.Mineau, A.Fall (eds.) Proceedings of the
International Symposium on Knowledge
Representation, Use, and Storage Efficiency.
Simon Fraser University, Vancouver 1997, 2--13
also in W.Gaul, H.Locarek-Junge (Eds.)
Classification in the Information Age. Springer,
Berlin-Heidelberg 1999, 344--356.
14
Concepts, Judgments and Conclusions
Dau, Eklund and Wille (2005)
24 I. Kant Logic. Dover, New York 1988.
15
Diagrammatic Reasoning why line diagrams are a
good idea

• In human reasoning, information is obtained from
  several modalities sentences, diagrams, moving
  pictures, etc. may be involved.
• Shin 34 writes, Recognizing the actual
  practice of this multi-modal reasoning,
  researchers have started focusing on multi-modal,
  or heterogeneous, representation systems, which
  employ both symbolic and diagrammatic elements.

34 S. J. Shin The Iconic Logic of Peirces
Graphs. Bradford Book, Massachusetts, 2002.
16
Habermas Concept of Society and its relationship
to FCA
Objects
OBJECTIVE PHYSICAL WORLD
NORMATIVE SOCIALWORLD
SUBJECTIVE PERSONAL WORLD
THE LIFE WORLD
Attributes
Incidence
17
Habermas Concept of Society and its relationship
to FCA
Context
OBJECTIVE PHYSICAL WORLD
NORMATIVE SOCIALWORLD
SUBJECTIVE PERSONAL WORLD
THE LIFE WORLD
Scales
Interpretation
18
Habermas Concept of Society and its relationship
to Peirces pragmatism
Language
OBJECTIVE PHYSICAL WORLD
NORMATIVE SOCIALWORLD
SUBJECTIVE PERSONAL WORLD
THE LIFE WORLD
Discourse
Argumentation
19
Habermas Concept of Society and its relationship
to Peirces pragmatism
Concepts
OBJECTIVE PHYSICAL WORLD
NORMATIVE SOCIALWORLD
SUBJECTIVE PERSONAL WORLD
THE LIFE WORLD
Judgements
Conclusions
20
Habermas Concept of Society and its relationship
to Peirces pragmatism
Firstness
OBJECTIVE PHYSICAL WORLD
NORMATIVE SOCIALWORLD
SUBJECTIVE PERSONAL WORLD
THE LIFE WORLD
Secondness
Thirdness
21
Reciprocal Discourse

• How do we know that CKP complies with the TCA?
• We test CKP by determining whether TCA validity
  holds via the following 4 rules.

• The statement communicated is true w.r.t. the
  objective world
• The statement is right w.r.t. the normative world
• The statement is honest w.r.t. the speakers
  subjective world
• The statement is comprehensible

Habermas, 1984, p. 99
22
Conceptual Knowledge Processing

• The adjective Conceptual in the name
  Conceptual Knowledge Processing underlines the
  constitutive role of the thinking, arguing and
  communicating human being for knowledge and its
  processing.
• The term Processing refers to the process in
  which something is gained which may be knowledge
  or something approximating knowledge such as a
  forecast, an opinion, a casual reason etc.
• To process knowledge, formal elements of language
  and procedures must be activated. This
  pre-supposes formal representations of knowledge
  and, in turn, knowledge must be constituted from
  such representations by humans not machines

23
Relationship to a Research Program In Computer
and Information Science

• Applications-based ? knowledge/document
  management
• Specialized to Semantic Web Technologies
• Ontological ? human-centered, knowledge-based,
  deductive
• Data Driven ? concept clustering/machine
  learning/information retrieval

24
Conclusions

• The major positives were
• the line diagram is useful for conceptualising
  email within mailboxes
• users found it easier to locate emails using
  Mail-Strainer
• the tool is simple to operate and provides a high
  level of customisability in regards to Structure
  and Query Management
• the tool would be useful for someone who had many
  emails embedded within their Webmail mailboxes
• the visual design of each of the main
  Mail-Strainer components is appealing.

• The major negatives were
• learning curve associated with reading the line
  diagram
• the Concept Lattice is invaluable but user needs
  more assistance
• a monster lattice was a reoccurring issue and
  affected the overall performance of the
  Mail-Strainer (and therefore SquirrelMail), this
  affected user judgement in relation to reading
  the line diagram.
• Some people had trouble identifying certain
  buttons and features within the Concept Lattice
  and Navigational Panel

25
Key References

• Formal Concept Analysis applications to
  Requirements Engineering and Design, T. Tilley,
  PhD Thesis, The University of Queensland, 2004.
• A Survey of Formal Concept Analysis Support for
  Software Engineering Activities, T. Tilley, R.
  Cole, P. Eklund, P. Becker, Formal Concept
  Analysis State of the Art, Springer Verlag,
  2004.
• Concept Data Theory and Applications, C.
  Carpineto and G. Romano, Wiley, 2004.
• Formal Concept Analysis Mathematical
  Foundations, B. Ganter and R. Wille, Springer
  Verlag, 1999.
• Scalability in Formal Concept Analysis, R. Cole
  and P. Eklund, Computational Intelligence, 15(1)
  11-27, 1999.
• A Contextual-Logic Extension of TOSCANA, P.
  Eklund, B. Groh, G. Stumme and R. Wille, 7th Int.
  Conf on Conceptual Structures, pp. 453-467,
  Springer Verlag, LNCS 1867, 2000.
• Document Retrieval for Email Search and Discovery
  using Formal Concept Analysis, Journal of Applied
  AI, 17(3) 257-280, 2003.
• The Lattice of Concept Graphs of a Relationally
  Scaled Context. S. Prediger, R. Wille In W.
  Tepfenhart, W. Cyre (Eds.) Conceptual
  Structures Standards and Practices, Springer
  Verlag, Berlin New York 1999, 401-414.
• Protoconcept GraphsThe Lattice of Conceptual
  Contents Joachim H. Correia and J. Klinger, 2nd
  Int. Conf on Formal Concept Analysis, LNCS2961,
  pp. 17-28, 2004.
• The Logic System of Concept Graphs with Negations
  (and its Relationship to Predicate Logic), F.
  Dau, LNAI 2892, Springer Verlag, 2003.
• Introduction to Lattice and Order Theory, B.
  Davey and H. Preistly, Cambridge University
  Press, 2nd Edition, 2002.
• Information Visualization using Concept Lattices
  Can Novices Read Line Diagrams? P.W.Eklund,
  J.Ducrou, P.Brawn LNCS2961, pp.57-72, LNAI,
  Springer-Verlag, 2004
• Evaluation of Concept Lattices in a Web-Based
  Mail Browser Eklund and Domingo, Proc. 13th Int.
  Conf on Conceptual Structures, LNAI3596, pp.
  142-153, Springer, 2005
• D-Sift A Dynamic Simple Intuitive FCA Tool
  Ducrou, Wormuth and Eklund, Proc. 13th Int. Conf
  on Conceptual Structures, LNAI3596, pp. 142-153,
  Springer, 2005

• http//www.upriss.org.uk/fca/fca.html
• http//www.kvocentral.org
• http//toscanaj.sf.net
• http//tockit.sf.net
• http//www.mail-sleuth.com
• http//www.hiermail.com
• http//www.wormuth.info
• http//credo.fub.it
• http//www.uow.edu.au/peklund

26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
Concept Lattices
36
Current Systems Toscana(J)

• Toscana(J)
• Requires an FCA expert (and possibly a domain
  expert) to prepare the data a scale library in
  Siena/Anaconda
• Scales are realized via SQL queries to a DBMS
• Data analysis relies on this library of
  pre-defined scales which may be combined to
  produce nested line-diagrams

37
Current Systems Cernato

• Cernato
• Data preparation and analysis are not separated
• All data must be explicitly and singularly typed
  binary, enumerated, numeric, interval
• Database is internal
• Data exploration is iterative and dynamic,
  attribute by attribute
• A lower level of training is required for its
  users

38
D-Sift Aims

• Allow untrained users structural FCA analysis to
  their data
• Simplify the data preparation phase of the FCA
  workflow
• Provide a bureau-like FCA Web-service
• Iterative browsing, search and data exploration
  D-Sifts main emphasis is on data exploration and
  interaction
• Re-consider the Toscana(J) workflow

39
Known problems Complications of Dynamic Lattices

• Learning curve for reading lattice structures
• Eklund et al 15, found that untrained users
  could understand concept lattices but only after
  some instruction.
• The Monster Lattice problem
• Domingo et al 16, found that users tend to
  continually add attributes until immense
  unreadable results were created.
• Data Understanding
• If a user does has no knowledge of the underlying
  data, then reading of results can be difficult.

40
UnConstrained FCA Tools Warp9/FCA
Scalability in Formal Concept, Cole and Eklund,
Computational Intelligence, Vol. 15(1) pp. 11-27,
1999.
41
From CEM A Conceptual Email Manager, Cole and
Stumme, 7th International Conference on
Conceptual Structures, ICCS'2000 and Journal of
Applied AI, 2003.
42
Mail-Sleuth

• MS Outlook
• Flash interface
• Stylised interface
• Experiments with new forms of lattice
  visualisation
• No nested line diagrams

• Email Analysis (ICCS2003)

43
Docco (Becker Roberts et al.) part of Tockit
http//tockit.sf.net
Tockit was established by the DFG/ARC Irex funded
collaboration, A Framework for Conceptual
Knowledge Processing, Eklund and Wille,
1999/2000/2001
44
Mail-Strainer (Domingo and Eklund, 2004)
1. engineer a method for indexing mailboxes, via
IMAP, and determine how to integrate the
Mail-Strainer framework into opensource 2. adapt
Mail-Sleuth to a Web-based context, examining the
index structures built in 1. 3. create the
necessary navigational structures required to
mimic Mail- Sleuth
B

• Domingo and Eklund (ICCS2005)

45
IRIX C STL
Specialised Analysis using FCA Software
Engineering
Sorted/Unsorted Hash/non-hash Single/Multiple
46
Conclusions

• participants were able to understand and
  communicate the role of the 4 main components
  the Concept Lattice, Navigation Panel,
  Mail-Strainer Tree and Folder Manager.