1 A collaborative problem solving method, using terms classification Samuel BASSETTO, Ali SIADAT, Patrick MARTIN (speaker)

1

A collaborative problem solving A collaborative problem solving method, using terms classificationmethod, using terms classification

Samuel BASSETTO, Ali SIADAT, Samuel BASSETTO, Ali SIADAT, Patrick MARTIN (speaker)Patrick MARTIN (speaker)

22

1.1. What we want to do…proposition of a What we want to do…proposition of a knowledge modelknowledge model

2.2. Test of this model… toward a structured solving Test of this model… toward a structured solving methodmethod

3.3. The study case and results The study case and results 4.4. Conclusion et discussionsConclusion et discussions

A collaborative problem solving method, using A collaborative problem solving method, using terms classificationterms classification

OU

TLIN

EO

UTL

INE

1234

33

From the text to a matrix : the model of expert’s analysis

The

knw

oled

ge m

odel

The

knw

oled

ge m

odel

A scientometric approach

Analysis 1

Text of the analysis of

expert number 1

…Analysis 2

Text of the analysis of

expert number 2…

Analysis 3Text of the analysis of

expert number 3…

(KW1,KW2,KW3,KW4)

(KW2,KW5,KW6,KW1)

(KW4,KW3,KW2,KW6)

KW1 KW2 KW3 KW4 KW5 KW6 Analysis 1 1 1 1 1 Analysis 2 1 1 1 1 Analysis 3 1 1 1 1

Key Word base

Instance of analysis

1234

= A

[7]

44

A distance between analysis…Th

e kn

owle

dge

mod

elTh

e kn

owle

dge

mod

el

Calulation of a distance (named Δ) between analysis

KW1 KW2 KW3 KW4 KW5 KW6 Analysis 1 1 1 1 1 Analysis 2 1 1 1 1 Analysis 3 1 1 1 1

Key word base

Instance of analysis

KW1 KW2 KW3 KW4 KW5 KW6 Analysis 1 A11 A12 A13 A14 A14 A16 Analysis 2 A21 A22 A23 A24 A25 A26 Analysis 3 A31 A32 A33 A34 A35 A36

Generic notation

Aij = the value of the ith analysis for the jth key word

)(*)(

)())()((

2

kAnalysisiAnalysis

AikAijkAnalysisiAnalysis i

)*

*cos(),(

22

kj

j

AikAij

AkjAijAki

Lafourcade’s distanceGeneric distance

1234

= A

A =

[11]

55

Main issues of this seducing knowledge model

The

know

ledg

e m

odel

The

know

ledg

e m

odel

How to define a stable (from a set of analyses to an other one) and shared (between experts) Key Word Base ? – How to cope with evolution of terms from an analysis to its successor– How to cope with quiproquo (a key word seems to be used by an

expert in a particular way, but it means something else for this expert)– How to cope with polysemy (multiple meening of one word ex: table

for a informatician and for carpenter)– How to build at least a locally shared ontology ?

How to define the semantic linked at key word’s weight– What does +1, 0, 0.5… means– Can we authorize positive and negative weights ?

How to formulate the issue for minimizing impact on key word use and weights ?

1234

66

The

know

ledg

e m

odel

The

know

ledg

e m

odel

Main opportunities of this seducing model

Go a step further in the modelization of expert’s knowledge Ability to deep the knowledge acquisition process Formal use of codesign space (with key word) Ability to compare analysis (evolution of 1 analysis and

distance between analysis) Abitlity to build a case base

1234

[2],[3],[4]

77





OU

TLIN

EO

UTL

INE

1234

88

The

stru

ctur

ed m

etho

dTh

e st

ruct

ured

met

hod

The structured problem solving method

Application: the team problem solving method used for resolution of scraps happened in semiconductor’s industries (TOPS methods, 8D methods)

Operationnal goal of this method: reduce problem solving cycle time

Research goal: interact with expert’s knowledge

1234

99

The

stru

ctur

ed m

etho

dTh

e st

ruct

ured

met

hod

The structured problem solving method

The structure of the methodology: 5 steps (and some recursive ones) to go from the problem to the solution.

1234

Building the team

Analyse the problem for the first time

Restitute analyses for the first time

Meet to work toward the solution

Converge toward the solution

end Convergence achieved

start

Issue

1010

The

stru

ctur

ed m

etho

dTh

e st

ruct

ured

met

hod

Rules

The structure of the methodology: 5 steps (and some recursive ones) to go from the problem to the solution.

1234

Team Building

First problem analysis

First restitution

Working meetings

Convergence

end Convergence achieved

start

Issue End by the expression of expert’s analysis and sum up with a set of weighted key words

Definition of a shared vocabulary, used for key word : B0

Definition of a shared vocabulary, used for key word: B1

Use of B1 to sum up analysis

Retrieve of analysis matrix A in the co-design space

[12]

1111

The

stru

ctur

ed m

etho

dTh

e st

ruct

ured

met

hod

Convergence measure

Indicator I behavior:1) If the group converge: I -> 02) If the group diverge: I-> 13) If I->0 the group converge4) If I->1 the group diverge

1234 Mistake

in the paper : 3 & 4 have

been interverted

I -> 0 the group convergeI ->1 the group diverge

1212

The

stru

ctur

ed m

etho

dTh

e st

ruct

ured

met

hodIndicator calculation

Ir = Sin(Amax)Ie = Spread(Amax/3) I1 = Ie*Ir I2 = (Ie+Ir)/2

1234

I -> 0 the group convergeI ->1 the group diverge

))();((max ),( jAnalysisiAnalysisDMAXA PersonsPersonsji

)(1)( rCDrSpread Number of couples @ a distance of r

CD(r )= ________________________________Number of possibles couples

Number of couples @ a distance of rCD(r )= ________________________________

Number of possibles couples

Number of couples @ a distance of rCD(r )= ________________________________

Number of possibles couples

1313





OU

TLIN

EO

UTL

INE

1234

1414The

stud

y ca

seTh

e st

udy

caseStudy case presentation

Analysis Why the choice of a particular test banc of forged cog influence cog’s lifetime, then the measurement.

5 expert’s of forged cog and lifetime measurement (5 doctors engineers about the subject)

Test @ ENSAM laboratory

1234

1515The

stud

y ca

seTh

e st

udy

case

Results 3 Iterations of analysis B0: Instead of the technical aspect of the problem, experts

retrieve 36 differents Key Words B1: 17 alias have been build based on these 36 KW Use of codesign space by Alias Key words

1234

Indicator (Ie+Ir)/2

Indicator Ie*Ir

Iteration Number Value Value

1 1,00 1,00

2 0,85 0,21

3 0,44 0,00

I1I2

1616





OU

TLIN

EO

UTL

INE

1234

1717Conc

lusi

on a

nd d

iscu

ssio

nCo

nclu

sion

and

dis

cuss

ion

Conclusion Compared to other tests (not mentionned in this paper) has

lead @ a quantified results Technological problem help experts to focus on key word

without QUIPROQUO & POLYSEMY Building a shared terminology (group ontology) is required Necessary to define the signification of key word weight Indicators can help the group to manage weak signals

(persons who will not take the lead)

1234

(software design, software development post mortem

analysis)

1818Conc

lusi

on a

nd d

iscu

ssio

nCo

nclu

sion

and

dis

cuss

ion

Still open questions…1234

At least, 5 questions are still pending…– How to define the problem ?Ex: If the question is: Is the function f1 involved ? And a key word is f1 yes this

function is involved, the answer can be sum-up with (1,f1)At the opposite, if the question is: Is the function f1 non involved ?, the answer

will be, no, this function is involved and the answer can be sum up with (f1, -1) or (f1,0) !

– What is expressive power of weighted key word ? what does mean -1,0,1…– Can we take an ‘alien’ ontology as a key word base– Is it possible to reuse a key word index defined by another group ?– How the codesign space can be tooled so as better managing the key

word creation ?

1919Conc

lusi

on a

nd d

iscu

ssio

nCo

nclu

sion

and

dis

cuss

ion

Thank you for your attentionDanke, Gracie, Merci

Questions ???

1234

1 A collaborative problem solving method, using terms classification Samuel BASSETTO, Ali SIADAT, Patrick MARTIN (speaker)

Documents