• Aucun résultat trouvé

Formation de coalitions pour une composition de services Web fondée sur la confiance dans les réseaux sociaux

N/A
N/A
Protected

Academic year: 2021

Partager "Formation de coalitions pour une composition de services Web fondée sur la confiance dans les réseaux sociaux"

Copied!
27
0
0

Texte intégral

(1)

Formation de coalitions pour une composition de

services Web fond´

ee sur la confiance dans les r´

eseaux

sociaux

Amine Louati1 Joyce El Haddad 2 Suzanne Pinson2

1Institut mines T´el´ecom, T´el´ecom ´Ecole de Management, LITEM, France 2PSL, Universit´e Paris-Dauphine, LAMSADE CNRS UMR 7243, France

(2)

Outline

1 Motivation

2 Background

3 Trust model

4 Coalition formation process description

5 Experimental Results

6 Conclusion and Perspectives

(3)

Motivation

Outline

1 Motivation

2 Background

3 Trust model

4 Coalition formation process description

5 Experimental Results

(4)

Motivation

Motivation

Service Composition

Satisfies a complex user needswhich cannot be achieved by an atomic service

Allows the definition of value addedapplications, which have the potential to reduceeffort and time of development [Ponn 02]

(5)

Motivation

The model : a multi-agent model

Able to performcomplex and distributed tasks

Support different forms of interactionincluding negotiation and coordination

Capable to extract and interpret information

Existing multi-agent approaches

Planning (graph, state, actions) [Paik 06, Ponn 02, Siri 04, Tong 11, Xu 11]

Coordination (reasoning, roles, negotiation) [Charif 13, Siala 11, Wang 06, Maam 05]

Cooperation (organization, preferences, sociability) [Grif 03, Ermo 03, Mull 06, Hong 09, Bour 09]

(6)

Motivation

Proposition : a coalition formation process

Challenges

How to integrate social dimension in the coalition process ⇒ Social trust model

How to ensure providers autonomy to decide with whom to cooperate ⇒ Endow agents with the ability to participate in the coalition process according to theirpreferences

How to enable agents to leave coalition when they are not satisfied ⇒ Definition of anincremental, dynamic and overlapping protocol for member selection

(7)

Background

Outline

1 Motivation

2 Background

3 Trust model

4 Coalition formation process description

5 Experimental Results

(8)

Background

Concepts Definitions

Social network : Given a set A = {a1, a2, ..., al} of agents and a set

E ⊆ A × A of edges, a social network is a connected graph

G =< A, E > where an edge (ak, aj) ∈ E represents an asymmetric

trust relationship between ak and aj

Agent: An agent ak ∈ A is an autonomous entity such that

ak =< Sk, Trust, ET , CT , λInfk, λSupk, βk, Blistk >

Service: A service s is a tuple such as s = (in, out, f , q1, q2, q3)

Louati, A., El Haddad, J., Pinson, S.

A Multilevel Agent-based Approach for Trustworthy Service Selection in Social Networks in IAT 2014 3 a a6 a8 a9 10 a 1 a 2 a a5 4 a 11 a a7 12 a requester i a agent

(9)

Background

Concepts Definitions

User query : Let F be the definition domain of available

functionalities. A user query Q = {f1, f2, . . . , fn| ∀1 ≤ i ≤ n, fi ∈ F } is

a finite set of functionalities

Coalition: Let Q be a user query. A coalition

c = {(f1, x1), . . . , (fi, xi), . . . , (fn, xn) | ∀i ∈ [1, n], ∃k ∈ [1, s] such as

xi = ak et ak ∈ Ai} is a set of agents that satisfy Q.

Q={f1, f2, f3} 2 a 4 a 12 a c1

(10)

Background

Multi-agent model

Candidate (provider agent) Broker (requester agent) Member (provider agent) Trust model Trust model Trust model User query

Figure –A broker-based multi-agent model for dynamic service composition

Agents cooperate to satisfy complex user’s needs based on decentralized decision makingguided by trust in cooperation

(11)

Trust model

Outline

1 Motivation

2 Background

3 Trust model

4 Coalition formation process description

5 Experimental Results

(12)

Trust model

Trust

Trust in cooperation CT (ak, aj) = ( 1 if NbSollk[j ] = 0 NbMemk[j ] NbSollk[j ] otherwise

Level of reliability of a candidate aj according to a member ak based on

their history of cooperation

Trust in coalition

evalC (ak, cz) =

P

at∈czCT (ak, at)

|cz|

Degree of satisfaction of a candidate ak to join a coalition cz

(13)

Coalition formation process description

Outline

1 Motivation

2 Background

3 Trust model

4 Coalition formation process description

5 Experimental Results

(14)

Coalition formation process description

Service discovery and selection process

(15)

Coalition formation process description

(16)

Coalition formation process description

Service discovery and selection process

(17)

Coalition formation process description

(18)

Coalition formation process description

Coalition formation process description

Sequentialprocess composed of three phases :

1 Initial Coalitions Generation (e.g. C = {c1 = {a2}, c2= {a6}})

2 Member Selection

3 Best coalition choice c ← Argmax 1≤z≤|C|

P

at ∈czET (at)

|cz|

(19)

Coalition formation process description

Coalition formation process description

Sequentialprocess composed of three phases :

1 Initial Coalitions Generation (e.g. C = {c1 = {a2}, c2= {a6}}) 2 Member Selection

3 Best coalition choice c ← Argmax 1≤z≤|C|

P

at ∈czET (at)

(20)

Coalition formation process description

Coalition formation process description

Sequentialprocess composed of three phases :

1 Initial Coalitions Generation (e.g. C = {c1 = {a2}, c2= {a6}}) 2 Member Selection

3 Best coalition choice c ← Argmax 1≤z≤|C|

P

at ∈czET (at)

|cz|

(21)

Experimental Results

Outline

1 Motivation

2 Background

3 Trust model

4 Coalition formation process description

5 Experimental Results

(22)

Experimental Results

Experimental methodology

```` ```` ```` `` Configuration Scenario A B C D E User query 2 3 4 5 6 services 5 5 5 5 5 timeout (ms) 3000 3000 3000 3000 3000

Table –Definition of test scenarios

(23)

Experimental Results

Performance of the coalition process

A B C D E 0 % 20 % 40 % 60 % 80 % 100 % 100100100 10098 100100 100100 100100 89 93 88 67 58 67 71 57 46 0 0 27 10 18 0 0 0 0 3 Test scenario P ercentage of generated coalitions

5 coalitions 4 coalitions 3 coalitions 2 coalitions 1 coalition 0 coalition

Figure –Percentage of generated

coalitions per test scenario

A B C D E 0 2 4 6 0 0 0.6 1.6 4.45 Test scenario Average frequency of abandon

Figure –The average frequency of

(24)

Conclusion and Perspectives

Outline

1 Motivation

2 Background

3 Trust model

4 Coalition formation process description

5 Experimental Results

6 Conclusion and Perspectives

(25)

Conclusion and Perspectives

Conclusion

Trust-based dynamic coalition formation process for service composition in social networks

Incremental, dynamic and overlapping selection protocol

Agent cooperate based on a decentralized decision-making process guided by trust in cooperation

(26)

Conclusion and Perspectives

Perspectives

More experimentation

Investigate the correlation between the quality of the chosen composite service and the trustworthiness of its members

Analyze the impact of the variation of the maximum layer value on the coalition formation process

Integrate negotiation in the coalition formation process : persuasions strategies

Examine the coalition stability : parallel coalition generation and define the utility of a service composition

(27)

Conclusion and Perspectives

Thank you !

Questions ... ?

Références

Documents relatifs

En termes de leçons tirées, nous avons pu spécialiser les ontologies générales proposées, et nous utilisons ce cadre commun par exemple pour étendre la recherche

While the REA model describes very well the concept of control over the resources by economic agents, it is less clear how to model the two other aspects of coalition power

Comme l’algorithme Bucket, l’algorithme MiniCon fonctionne en deux ´ etapes qui sont la recherche des vues pertinentes pour la r´ e´ ecriture de la requˆ ete et la combinaison de

step response of a zero-phase filter (Butterworth order 8 applied using filtfilt), cutoff as indicated in legend. B: response to a pulse of a Butterworth order 8 filter, cutoff

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des

We model community detection as a coalition formation game in which individuals in a social network are modelled as rational players aiming to improve the

Le mucus va entrainer avec lui des cellules mucineuses qui vont essaimer dans la cavité péritonéale et se multiplier pour former un pseudomyxome. Pour cela

Regarding the rain effects corrections, we found significant changes for pairs PUS-KLA, KLA-PLA, and DEL-PLA, whereas, due to their low correlation with GWL curve, the