Social Computational Trust Model (SCTM): A Framework to ... · 0.7 0.8 0.9 The value of benevolence...
Transcript of Social Computational Trust Model (SCTM): A Framework to ... · 0.7 0.8 0.9 The value of benevolence...
Social Computational Trust Model (SCTM): A Framework to Facilitate the
Selection of Partners
Ameneh DeljooSystems and Networking Lab
University of Amsterdam [email protected]
Motivation
vNetwork of organizations evolve over time and become more complex,
vFind a “right” partner is a challenging task
We need to: vDefine a more sophisticated and
computationally executable method to select the “right" partner for sharing data and intelligence.
Contributions
vThe Social Computational Trust Model (SCTM) represents social trust and its components, which are important for evaluating the partners.
v Risk assessment through the SCTM model. The SCTM facilitates risk-based partner selection to select the “right" partner to collaborate in joint tasks.
Trust and its Antecedents
v“x” expects “y” to do task (") and “y” will not exploit vulnerabilities of “x” when “y” faced with the opportunity to do so. Therefore, “y”:
vHas the potential ability to perform a given task (competence),vAdheres to a set of rules agreed upon and acts accordingly to fulfill the commitments
(integrity), andvActs and does good even if unexpected contingencies arise (benevolence).
kj
Competence
Integrity
Benovelence
Trustworthiness
Trustee
Trustor
Trust OutCome >= Expectation Risk Assessment
ContextA Set of
Principles
Kindship
Evidence* Current (S), Context
Current (S), Context Evidence*
Evidence
Adopted from Mayer et al. (1995) ``An Integrative Model of Organizational Trust"
Social Computational Trust Model (SCTM)
vIdentify two distinctive trustworthiness factors (Benevolence and Competence)
vEvaluate Trust in a dynamic wayvGather the direct and indirect evidence on a trustee vUpdate Trust value
1 Integrity has been considered as a part of Benevolence function.
benevolenceEvaluation Function
Ben (x,y, si )
Trustworthiness Evaluation Function
TW (x,y, si )Trust
Tr ( x,y, si )
Ed(x, y, si ; Kbx )
Ec(x, y, si )
Ben (x,y, si )
TW (x,y, si )
competenceEvaluation Function
Com (nbry ,y, si )
Com (nbry , y, si )
Kbx
Outcome of a taskTask typeReptReqtDestination’s IdOriginator’s
Id
Notation
1Dimensions are: d1 = trustor, d2= trustee , d3 = time, d4= location, d5= task, d6=complexity, d7= deadline, d8= Outcome
1
In order to define the situations that lead to an agreement between a trustor and a trustee: vd1 = trustor, vd2= trustee , vd3 = time, vd4= location,vd5= task,vd6=complexity, vd7= deadline, vd8= Outcome vThree different outcome of tasks
• !" !#$$%&$ "#'(• !"" !#$$%&$ "#'( )&'ℎ "+$,(• - -&.$,'+
Dimensions
val ("/) = 11 , &% "/= !"0.5 , &% "/= !""0 , &% "/= -
val (!") = $1 , '( !"= )!0.5 , '( !"= )!!0 , '( !"= -
Calculate the Outcome
benevolenceEvaluation Function
Ben (x,y, si )
Trustworthiness Evaluation Function
TW (x,y, si )Trust
Tr ( x,y, si )
Ed(x, y, si ; Kbx )
Ec(x, y, si )
Ben (x,y, si )
TW (x,y, si )
competenceEvaluation Function
Com (nbry ,y, si )
Com (nbry , y, si )
Kbx
Outcome of a taskTask typeReptReqtDestination’s IdOriginator’s
Id
vd8= Outcome vThree different outcome of tasks
)! ).//('/ !.01)!! ).//('/ !.01 2'0ℎ !4/51
- -'6/504
Evidence Gathering: Direct evidence
vA trustor looks at its Kb to collect the evidence on a trustee based on past interactions.
!"#$ . ⟶ [0,1]
,-(/, 0, 12; 456) = {-:(x, y,12) ∈ 456}!"#$ ,-(/, 0, 12; 456) = =
>?∑$A(6,B,CD)∈E$(6,B,CD; FG?) !"# -:(x, y,12)
val (-:) = H1 , IJ -:= K-0.5 , IJ -:= K--0 , IJ -:= M
,N6 = OPQ5RS TJ ROSUIR1 IO UℎR W5X1
Z
WD
A
Y
X
C
B
M
N
Request the evidence
Ed (X,Y)
Evidence Gathering: Indirect evidence
vA trustor asks a trustee’s direct neighbors to send him their evidence on a given trustee.
!"#$ . ⟶ [0,1]
Ec (./01, y,34) = { Ed(u, y,34;9/:) | < ∈ ./01}!"#$ ?@(A, B, 34) =
E
FGHI∑KL(:,1,MN; PQR)∈K$(SQTU,1, MN) !"#L ?V(u, y,34; 9/:)
Z
WD
A
Y
X
C
B
M
N
Request the evidence
Ec (W,Y)
Ec (Z,Y)
Ec (A,Y)
Ec (M,Y)
WSQT = .<X/Y0 Z[ .Y\]ℎ/Z03 _ℎ"_ @Z._0\/<_Y _Z _ℎY !"#$
Z
WD
A
Y
X
C
B
M
N
Request the evidence
Ed (X,Y)
Benevolence Function
vBased on the direct interactions between !"#$!%" & '() !"#$!** + in the situation $,.
-*( &, +, $, = 0'12 3)(&, +, $,, 567)
Competence Function
vEvaluate based on the all available evidence on Trustee (e.g. y,z)
!"# $%&', ), *+ = -./0 Ec($%&'3, y, *+) , $%&'3 = $%&'\{7}
Deljoo, Ameneh, et al. "The Impact of Competence and Benevolence in a Computational Model of Trust." IFIP International Conference on Trust Management. Springer, Cham, 2018.
Z
WD
A
Y
X
C
B
M
N
Request the evidence
Ec (W,Y)
Ec (Z,Y)
Ec (A,Y)
Ec (M,Y)
Estimating Trust1 based on Competence and Benevolence functions
!"($, &, '() =12 (-./(0123, &, '() + 560 $, &, '( )
1 Integrity has been considered as a part of Benevolence function.
!2 $, &, '( = !"($, &, '()
Risk Estimation
Risk Estimation
Interaction Risk ("# $, &, '# ) in the Alliance Consists of:
vRelational Risk ") $, &, '# : The probability and consequence of not having a successful cooperation.
vPerformance Risk ("+ $, &, '# ): The probability and consequences that alliance objectives are not realized despite satisfactory cooperation among the partner.
Propositions
Proposition1
Benevolent1 behavior of partners increases trust and reduces former perceived relational risk in the alliance.
!" #, %, &' ∝ (1 − ,-. #, %, &' )
Proposition 2
The perceived performance risk will be reduced if the competence of the given member is high.
!0 #, %, &' ∝ 1 − 123 .456, %, &'1Some of the scholars consider faith and good intentions instead of benevolence.
Interaction risk
InteractionRiskisgivenby:
34 5, 7, 84 = 3: 5, 7, 84 + 3< 5, 7, 84
34 5, 7, 84 = =>(1 − BCD 5, 7; 84 ) + =G 1 − HIJ 5, 7; 84
34 5, 7, 84 = K 1 − BCD JLMN, 7, 84 + 1 − α 1 − HIJ 5, 7, 84 , 0 ≤ K ≤ 1
Perceived Interaction Risk( Ri ( x,y, Si ))
Benevolence Function
Risk Estimation
Competence Function
Relation Risk (Rr ( x,y, Si ) ) Performance Risk (Rp ( x,y, Si ))
W1W2
=> = K , =G= 1 − KT. Das, B.-S. Teng, Risk types and inter-rm alliance structures, Journal of management studies 33 (6) (1996) 827{843.
Case Study
Z
WD
A
Y
X
C
B
M
N
A Collaborative Network
Simulation settings and their illustrations
Z
WD
A
Y
X
C
B
M
N
Scenario
Domain “N” wants to choose ideal domains for collaboration in order to mitigate and defend against a certain attack.
Task (!): Mitigate and defend against a certain attack.
Sub-tasks: v!"#: provide resources within a certain time window,v!"$: monitor a certain traffic,v!"%: block a certain link, v!"&: implement a certain counter measurement.
Selecting a “right” partner algorithm
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Com
pete
nce,
Ben
evol
ence
, Ri
s1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1s2
Domains
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1s3
Domains
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1s4
Com
pete
nce,
Ben
evol
ence
, Ri
T
TT
T
Competence
Benevolence
Ri
Competence
Benevolence
Ri
Z
ZM YAX X
XX A M Y
A
A
M
M
Y
Y
Z
Z
Result
Evaluation
v Epinion1 dataset a popular product review site. vEach user gives a trust value (–1 to 1) on other users.vAnd gives feedback ratings (1 to 5) on entities/items.
vV = 1, Fdd = 2 and Fd = 3; 4; 5.
vSelect five items from the dataset and evaluate benevolence and competence of
each item.
v SELCSP Algorithm and SOLUM Algorithm.
1http://www.trustlet.org/epinions.html
Evaluation Result
Members0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
The
valu
e of
ben
evol
ence
SELCSPSCTMSOLUM
X Y M Z A XMembers
0.4
0.6
0.8
0.9
The
valu
e of
com
pete
nce
SELCSPSCTMSOLUM
YM AZ
The value of benevolence for three different algorithms The value of competence for three different algorithms
Conclusion
vTo evaluate the trustworthiness of a trustee the direct and indirect evidence on the given trustee were taken into account.
vThe trust value is computed by two trust factors, namely competence and benevolence.
v Benevolence is computed from direct evidence between a trustee and a trustor vCompetence is assessed on the base of the received feedback from the other
alliance members (a trustee's direct neighbors). vWe are able to collect a variety of evidence on a trustee by introducing eight
dimensions for each context.
Conclusion
vThe interaction risk estimated through the SCTM by combining benevolence and competence.
v The weighting factors used to determine different weights to define the main trust factors in different trusting scenarios.
v We have shown that the stability of the alliance is dependent on the value of benevolence that led to a lower interaction risk.
vWe demonstrated that the SCTM is able to obtain comparable results to the other trust models that we evaluated.