Quan'ta've Argumenta'on Debates A. Rago 1 , F. Toni 1 , M. - - PowerPoint PPT Presentation

▶
quan ta ve argumenta on debates
SMART_READER_LITE
LIVE PREVIEW

Quan'ta've Argumenta'on Debates A. Rago 1 , F. Toni 1 , M. - - PowerPoint PPT Presentation

Mar 19, 2024 319 likes •972 views

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates A. Rago 1 , F. Toni 1 , M. Aurisicchio 1 & P. Baroni 2 1. Imperial College London 2. Universita` degli Studi di Brescia Cardiff Argumenta'on Forum 6 th July 2016

slide-1
SLIDE 1

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago 1, F. Toni 1, M. Aurisicchio 1 & P. Baroni 2
  • 1. Imperial College London
  • 2. Universita` degli Studi di Brescia

Cardiff Argumenta'on Forum

6th July 2016

1 / 24

slide-2
SLIDE 2

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Presenta'on Overview

1. Background

  • IBIS Charts and QuAD Frameworks
  • QuAD Algorithm/Seman'cs

2. Research Summary

  • Mo'va'on for DF-QuAD
  • DF-QuAD Algorithm
  • Comparison of the DF-QuAD and QuAD Algorithms
  • Proper'es Not Held by QuAD
  • Proper'es Shared with QuAD
  • Rela'onship to Abstract Argumenta'on
  • Reverse Engineering Func'onality
  • Applica'ons of QuAD Frameworks

3. Future Work 4. Conclusions

2 / 24

slide-3
SLIDE 3

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • IBIS (Issue Based Informa'on System) charts [Kunz and Ri_el, 1970].
  • QuAD (Quan'ta've Argumenta'on Debate) frameworks [Baroni et al. 2015].

– Special types of IBIS trees with base scores for nodes.

Background – IBIS Charts & QuAD Frameworks

3 / 24

QuAD Framework [www.arganddec.com]

slide-4
SLIDE 4

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • IBIS (Issue Based Informa'on System) charts [Kunz and Ri_el, 1970].
  • QuAD (Quan'ta've Argumenta'on Debate) frameworks [Baroni et al. 2015].

– Special types of IBIS trees with base scores for nodes.

  • Correspond to BAFs (Bipolar Argumenta'on Frameworks) [Cayrol and Lagasquie-Schiex, 2005].

Background – IBIS Charts & QuAD Frameworks

3 / 24

QuAD Framework [www.arganddec.com]

C4 P3 C1 A1

+

  • A2

P1 P2 C2 C3

+ +

BAF Framework

slide-5
SLIDE 5

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • Base scores are used by the QuAD algorithm to calculate each node’s overall strength.
  • Base scores and strengths are in [0,1].
  • Strength is a form of gradual acceptance [Cayrol and Lagasquie-Schiex, 2005].

Background – QuAD Algorithm/Seman'cs

4 / 24

slide-6
SLIDE 6

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • Base scores are used by the QuAD algorithm to calculate each node’s overall strength.
  • Base scores and strengths are in [0,1].
  • Strength is a form of gradual acceptance [Cayrol and Lagasquie-Schiex, 2005].
  • Firstly, a_acking and suppor'ng components (va and vs) are calculated for each node.

Background – QuAD Algorithm/Seman'cs

4 / 24

slide-7
SLIDE 7

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • Base scores are used by the QuAD algorithm to calculate each node’s overall strength.
  • Base scores and strengths are in [0,1].
  • Strength is a form of gradual acceptance [Cayrol and Lagasquie-Schiex, 2005].
  • Firstly, a_acking and suppor'ng components (va and vs) are calculated for each node.

Background – QuAD Algorithm/Seman'cs

4 / 24

slide-8
SLIDE 8

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • Base scores are used by the QuAD algorithm to calculate each node’s overall strength.
  • Base scores and strengths are in [0,1].
  • Strength is a form of gradual acceptance [Cayrol and Lagasquie-Schiex, 2005].
  • Firstly, a_acking and suppor'ng components (va and vs) are calculated for each node.

Background – QuAD Algorithm/Seman'cs

4 / 24

slide-9
SLIDE 9

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Background – QuAD Algorithm/Seman'cs

  • Recursive formulae are used for a_acking and suppor'ng components.

1

5 / 24

slide-10
SLIDE 10

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Background – QuAD Algorithm/Seman'cs

  • Recursive formulae are used for a_acking and suppor'ng components.

1

5 / 24

slide-11
SLIDE 11

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Background – QuAD Algorithm/Seman'cs

  • Recursive formulae are used for a_acking and suppor'ng components.

1

5 / 24

slide-12
SLIDE 12

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Background – QuAD Algorithm/Seman'cs

  • Recursive formulae are used for a_acking and suppor'ng components.
  • If the set of a_acker/supporter strengths is {} or a set of zeros it is considered ineffec0ve.

1

5 / 24

slide-13
SLIDE 13

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Background – QuAD Algorithm/Seman'cs

  • Recursive formulae are used for a_acking and suppor'ng components.
  • If the set of a_acker/supporter strengths is {} or a set of zeros it is considered ineffec0ve.
  • The aggrega'ng func'on then determines the strength in the [0,1] range:

1

5 / 24

va vs Strength Effec've Ineffec've va Ineffec've Effec've vs Ineffec've Ineffec've v0 Effec've Effec've (va + vs) / 2

slide-14
SLIDE 14

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Mo'va'on for DF-QuAD

  • Engineering Design selng:

– Issue – Which is the best method for controlling the ven'la'on of a dining room? – Answer 1 – Building management control – Answer 2 – User control

6 / 24

slide-15
SLIDE 15

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Mo'va'on for DF-QuAD

  • Engineering Design selng:

– Issue – Which is the best method for controlling the ven'la'on of a dining room? – Answer 1 – Building management control – Answer 2 – User control

  • Pro arguments are added at Stage 1:

– Pro 1 – Energy is saved – Pro 2 – Elderly occupants require more simple selngs – Pro 3 – User sa'sfac'on is increased

6 / 24

slide-16
SLIDE 16

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Mo'va'on for DF-QuAD

  • Engineering Design selng:

– Issue – Which is the best method for controlling the ven'la'on of a dining room? – Answer 1 – Building management control – Answer 2 – User control

  • Pro arguments are added at Stage 1:

– Pro 1 – Energy is saved – Pro 2 – Elderly occupants require more simple selngs – Pro 3 – User sa'sfac'on is increased

  • QuAD Algorithm:

Answer Strength at Stage 1 A1 0.925 A2 0.950

6 / 24

slide-17
SLIDE 17

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Mo'va'on for DF-QuAD

  • At Stage 2, a con argument a_acking A2 is then added:

– Con 1 – User negligence can lead to losses

7 / 24

slide-18
SLIDE 18

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Mo'va'on for DF-QuAD

  • At Stage 2, a con argument a_acking A2 is then added:

– Con 1 – User negligence can lead to losses

  • QuAD Algorithm:

Answer Strength at Stage 1 Strength at Stage 2 A1 0.925 0.925 A2 0.950 0.675

7 / 24

slide-19
SLIDE 19

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Mo'va'on for DF-QuAD

  • At Stage 2, a con argument a_acking A2 is then added:

– Con 1 – User negligence can lead to losses

  • QuAD Algorithm:
  • Large drop in A2’s strength between Stage 1 and 2 is dispropor'onate in some selngs, e.g.

Engineering Design.

  • In other selngs, e.g. E-Democracy, this may not be the case.

Answer Strength at Stage 1 Strength at Stage 2 A1 0.925 0.925 A2 0.950 0.675

7 / 24

slide-20
SLIDE 20

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • The aggrega'ng func'on’s range in one case is a subset of [0,1]:

Research Summary – Mo'va'on for DF-QuAD

8 / 24

  • therwise

va and vs effec've

slide-21
SLIDE 21

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • The aggrega'ng func'on’s range in one case is a subset of [0,1]:

Research Summary – Mo'va'on for DF-QuAD

8 / 24

  • therwise

va and vs effec've Strength

  • f A2 at

Stage 1

slide-22
SLIDE 22

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • The aggrega'ng func'on’s range in one case is a subset of [0,1]:
  • Discon'nuity, leading to counter-intui've behaviour in some applica'ons.

Research Summary – Mo'va'on for DF-QuAD

8 / 24

  • therwise

va and vs effec've Strength

  • f A2 at

Stage 1 Strength

  • f A2 at

Stage 2

slide-23
SLIDE 23

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – DF-QuAD Algorithm

  • A new “discon'nuity-free” algorithm for QuAD frameworks (DF-QuAD).
  • Incorporates many of the same concepts as the QuAD algorithm.

– Base score and strength in [0,1].

9 / 24

slide-24
SLIDE 24

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – DF-QuAD Algorithm

  • A new “discon'nuity-free” algorithm for QuAD frameworks (DF-QuAD).
  • Incorporates many of the same concepts as the QuAD algorithm.

– Base score and strength in [0,1].

  • A single func'on used for both the a_acking and suppor'ng components.

9 / 24

slide-25
SLIDE 25

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – DF-QuAD Algorithm

  • A new “discon'nuity-free” algorithm for QuAD frameworks (DF-QuAD).
  • Incorporates many of the same concepts as the QuAD algorithm.

– Base score and strength in [0,1].

  • A single func'on used for both the a_acking and suppor'ng components.

9 / 24

slide-26
SLIDE 26

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – DF-QuAD Algorithm

  • A new “discon'nuity-free” algorithm for QuAD frameworks (DF-QuAD).
  • Incorporates many of the same concepts as the QuAD algorithm.

– Base score and strength in [0,1].

  • A single func'on used for both the a_acking and suppor'ng components.

9 / 24

slide-27
SLIDE 27

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • The func'on recursively calculates values for a_acking and suppor'ng components:

Research Summary – DF-QuAD Algorithm

10 / 24

slide-28
SLIDE 28

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • The func'on recursively calculates values for a_acking and suppor'ng components:

Research Summary – DF-QuAD Algorithm

10 / 24

slide-29
SLIDE 29

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • The func'on recursively calculates values for a_acking and suppor'ng components:

Research Summary – DF-QuAD Algorithm

10 / 24

slide-30
SLIDE 30

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • The func'on recursively calculates values for a_acking and suppor'ng components:
  • The combina'on func'on then uses the base score and the difference between the a_acking/

suppor'ng components to calculate the strength.

Research Summary – DF-QuAD Algorithm

10 / 24

slide-31
SLIDE 31

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni
  • The func'on recursively calculates values for a_acking and suppor'ng components:
  • The combina'on func'on then uses the base score and the difference between the a_acking/

suppor'ng components to calculate the strength.

  • The strength is not restricted to a subset of [0,1] when a_ackers and supporters are

effec've.

  • Iden'cal results to the QuAD algorithm when a_ackers or supporters are ineffec've.

Research Summary – DF-QuAD Algorithm

10 / 24

slide-32
SLIDE 32

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Comparison of the DF-QuAD and QuAD Algorithms

  • Engineering Design selng as used for QuAD.

QuAD Algorithm:

11 / 24

Answer Strength at Stage 1 Strength at Stage 2 A1 0.925 0.925 A2 0.950 0.675

slide-33
SLIDE 33

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Comparison of the DF-QuAD and QuAD Algorithms

  • Engineering Design selng as used for QuAD.

QuAD Algorithm: DF-QuAD Algorithm:

11 / 24

Answer Strength at Stage 1 Strength at Stage 2 A1 0.925 0.925 A2 0.950 0.850 Answer Strength at Stage 1 Strength at Stage 2 A1 0.925 0.925 A2 0.950 0.675

slide-34
SLIDE 34

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Comparison of the DF-QuAD and QuAD Algorithms

  • Engineering Design selng as used for QuAD.

QuAD Algorithm: DF-QuAD Algorithm:

11 / 24

Answer Strength at Stage 1 Strength at Stage 2 A1 0.925 0.925 A2 0.950 0.850 Answer Strength at Stage 1 Strength at Stage 2 A1 0.925 0.925 A2 0.950 0.675

slide-35
SLIDE 35

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Comparison of the DF-QuAD and QuAD Algorithms

  • When QuAD’s strength is plo_ed for a constant base score (0.5).

12 / 24

slide-36
SLIDE 36

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Comparison of the DF-QuAD and QuAD Algorithms

  • When QuAD’s strength is plo_ed for a constant base score (0.5).

12 / 24

Ineffec've A_ackers Ineffec've Supporters

slide-37
SLIDE 37

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Comparison of the DF-QuAD and QuAD Algorithms

  • When QuAD’s strength is plo_ed for a constant base score (0.5).

12 / 24

Effec've A_ackers & Supporters Ineffec've A_ackers Ineffec've Supporters

slide-38
SLIDE 38

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Comparison of the DF-QuAD and QuAD Algorithms

  • When QuAD’s strength is plo_ed for a constant base score (0.5).

12 / 24

Discon'nuity

slide-39
SLIDE 39

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Comparison of the DF-QuAD and QuAD Algorithms

  • DF-QuAD plo_ed for the same base score shows results without a discon'nuity.

13 / 24

slide-40
SLIDE 40

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Proper'es Not Held by QuAD

14 / 24

  • As an a_acker’s or a supporter’s strength approaches 0, the framework becomes equivalent

to a framework without that argument: – As the strength of the a_acker C1 approaches 0:

slide-41
SLIDE 41

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Proper'es Not Held by QuAD

  • As an a_acker’s or a supporter’s strength approaches 0, the framework becomes equivalent

to a framework without that argument: – As the strength of the a_acker C1 approaches 0:

~

14 / 24

– Similarly for supporters. – Gives DF-QuAD its discon'nuity-free aspect.

slide-42
SLIDE 42

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Proper'es Not Held by QuAD

  • In DF-QuAD, the following results in a strength of A2 which is less than its base score:

Condi0on Strength Argument’s a_acking component is larger than its suppor'ng component Less than or equal to the base score

15 / 24

Answer QuAD Strength DF-QuAD Strength A1 0.24 0.16

slide-43
SLIDE 43

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Proper'es Not Held by QuAD

  • In DF-QuAD, the following results in a strength of A2 which is equal to its base score:

Condi0on Strength Argument’s a_acking component is larger than its suppor'ng component Less than or equal to the base score Argument’s a_acking component is equal to its suppor'ng component Equal to the base score

16 / 24

Answer QuAD Strength DF-QuAD Strength A1 0.35 0.2

slide-44
SLIDE 44

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Proper'es Not Held by QuAD

  • In DF-QuAD, the following results in a strength of A2 which is greater than its base score:

Condi0on Strength Argument’s a_acking component is larger than its suppor'ng component Less than or equal to the base score Argument’s a_acking component is equal to its suppor'ng component Equal to the base score Argument’s a_acking component is smaller than its suppor'ng component Greater than or equal to the base score

17 / 24

Answer QuAD Strength DF-QuAD Strength A1 0.675 0.85

slide-45
SLIDE 45

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Proper'es Shared with QuAD

  • The order of the sequence of a_acking or suppor'ng arguments does not affect the strength:

18 / 24

Stage 1 Pro before Con Con before Pro

slide-46
SLIDE 46

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Proper'es Shared with QuAD

  • The order of the sequence of a_acking or suppor'ng arguments does not affect the strength:

Stage 1 Stage 2

18 / 24

Pro before Con Con before Pro

slide-47
SLIDE 47

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Proper'es Shared with QuAD

  • The order of the sequence of a_acking or suppor'ng arguments does not affect the strength:

Stage 1 Stage 2 Pro before Con Con before Pro

≡

18 / 24

slide-48
SLIDE 48

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Proper'es Shared with QuAD

  • The order of the sequence of a_acking or suppor'ng arguments does not affect the strength:

Stage 1 Stage 2 Pro before Con Con before Pro

≡

18 / 24

  • More intui've, comprehensive proper'es also hold, for example:

– An a_acker being added will not increase the strength – A supporter being added will not decrease the strength

slide-49
SLIDE 49

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Rela'onship to Abstract Argumenta'on

  • QuAD and DF-QuAD yield the same results when a_ackers or supporters are ineffec've.
  • Abstract Argumenta'on frameworks [Dung, 1995] can be mapped to QuAD frameworks without

supporters [Baroni et al. 2015].

19 / 24

slide-50
SLIDE 50

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Rela'onship to Abstract Argumenta'on

  • QuAD and DF-QuAD yield the same results when a_ackers or supporters are ineffec've.
  • Abstract Argumenta'on frameworks [Dung, 1995] can be mapped to QuAD frameworks without

supporters [Baroni et al. 2015].

19 / 24

A B C

  • QuAD Framework

AA Framework

slide-51
SLIDE 51

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Rela'onship to Abstract Argumenta'on

  • QuAD and DF-QuAD yield the same results when a_ackers or supporters are ineffec've.
  • Abstract Argumenta'on frameworks [Dung, 1995] can be mapped to QuAD frameworks without

supporters [Baroni et al. 2015].

  • If all base scores set to 1:

– Arguments in grounded extension have strength of 1. – Other arguments have strength of 0.

19 / 24

A B C

  • QuAD Framework

AA Framework

slide-52
SLIDE 52

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Reverse Engineering Func'onality

  • Reverse engineering: a user can engineer the framework to give a required ranking or score.

20 / 24

slide-53
SLIDE 53

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Reverse Engineering Func'onality

  • Reverse engineering: a user can engineer the framework to give a required ranking or score.
  • Engineering Design Example:

– Determine the BS(C1) at which the strength of A2 equals the strength of A1.

20 / 24

slide-54
SLIDE 54

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Reverse Engineering Func'onality

  • Reverse engineering: a user can engineer the framework to give a required ranking or score.
  • Engineering Design Example:

– Determine the BS(C1) at which the strength of A2 equals the strength of A1. – BS(C1) = 0.05

20 / 24

slide-55
SLIDE 55

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Reverse Engineering Func'onality

  • Reverse engineering: a user can engineer the framework to give a required ranking or score.
  • Engineering Design Example:

– Determine the BS(C1) at which the strength of A2 equals the strength of A1. – BS(C1) = 0.05

20 / 24

  • Other altera'ons include:

– Increasing the strength of arguments by increasing base scores, e.g. increasing BS(P3) or BS(A2). – Reducing the strength of arguments by adding a_ackers, e.g. to C1. – Increasing the strength of arguments by adding supporters, e.g. to P3 or A2.

  • Not possible in QuAD.
  • Not desirable in some applica'ons, e.g. where manipula'on could be a problem.
slide-56
SLIDE 56

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Applica'ons of QuAD Frameworks

  • Arg&Dec [www.arganddec.com]

– QuAD framework – QuAD automa'c evalua'on – Unique transla'on from graph to matrix form

21 / 24

slide-57
SLIDE 57

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Applica'ons of QuAD Frameworks

  • Arg&Dec [www.arganddec.com]

– QuAD framework – QuAD automa'c evalua'on – Unique transla'on from graph to matrix form

  • designVUE [www3.imperial.ac.uk/

designengineering/tools/designvue]

– Debate mapping using IBIS graphs – Engineering design specific – Less limita'ons than Arg&Dec – Qualita've weigh'ng measures

21 / 24

slide-58
SLIDE 58

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Research Summary – Applica'ons of QuAD Frameworks

  • Arg&Dec [www.arganddec.com]

– QuAD framework – QuAD automa'c evalua'on – Unique transla'on from graph to matrix form

  • designVUE [www3.imperial.ac.uk/

designengineering/tools/designvue]

– Debate mapping using IBIS charts – Engineering design specific – Less limita'ons than Arg&Dec – Qualita've weigh'ng measures

  • Other Applica'ons

– E-Democracy – Argument Mining – Medical Decision Support – Legal Reasoning

21 / 24

slide-59
SLIDE 59

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Future Work

  • Further development of the QuAD framework:

– Vo'ng – Comparison with other algorithms/seman'cs for QuAD frameworks – Addi'onal proper'es for algorithms [Amgoud & Ben-Naim, 2016]

22 / 24

slide-60
SLIDE 60

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Future Work

  • Further development of the QuAD framework:

– Vo'ng – Comparison with other algorithms/seman'cs for QuAD frameworks – Addi'onal proper'es for algorithms [Amgoud & Ben-Naim, 2016]

  • Reverse engineering

– Rela'onships to other altera'ons techniques, e.g. Enforcement [Bisquert & Cayrol, 2013]

22 / 24

slide-61
SLIDE 61

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Future Work

  • Further development of the QuAD framework:

– Vo'ng – Comparison with other algorithms/seman'cs for QuAD frameworks – Addi'onal proper'es for algorithms [Amgoud & Ben-Naim, 2016]

  • Reverse engineering

– Rela'onships to other altera'ons techniques, e.g. Enforcement [Bisquert & Cayrol, 2013]

  • Rela'onships with other frameworks:

– Matrix method [Aurisicchio et al. 2015] – Bipolar Argumenta'on Frameworks [Baroni et al. 2015; Amgoud et al. 2008] – Fuzzy Logic – Argumenta'on Labelling

22 / 24

slide-62
SLIDE 62

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

Conclusions

  • We have presented the DF-QuAD Algorithm:

– Automa'c evalua'on within QuAD frameworks – Discon'nuity-free algorithm – Shares many important proper'es with QuAD, and holds some new ones – Allows for reverse engineering

23 / 24

slide-63
SLIDE 63

Discon'nuity-Free Decision Support with Quan'ta've Argumenta'on Debates

  • A. Rago, F. Toni, M. Aurisicchio & P. Baroni

24 / 24

Thank You Any Ques'ons?