Fuzzy Consensus on QoS in Web Services ... - Semantic Scholar

Fuzzy Consensus on QoS in Web Services Discovery Wei-Li Lin1, Chi-Chun Lo1, Kuo-Ming Chao2, Muhammad Younas3 1

Institute of Information Management, National Chiao-Tung University, Hsin-Chu, 300, Taiwan [email protected], [email protected] 2 DSM Research Group, Department of Computer and Network Systems, Coventry University, Coventry, CV1 5FB, UK [email protected] 3 Department of Computing, Oxford Brookes University, Oxford, UK [email protected]

ABSTRACT Ensuring consensus on QoS attributes is imperative for the discovery of web services. Inconsistent views of service providers and consumers on QoS attributes could be problematic, even though they have consensus on characteristics. For instance, they may have distinct views of the QoS attribute of service reliability, wherein a consumer considers that a service is reliable if its success rate is 99.9% or 100%, while providers may consider its service as reliable if its success rate is higher than 90%. Different expectations in service discovery could hinder prevailing of the web services. This paper proposes a QoS Consensus Moderation Approach (QCMA) in order to perform QoS consensus and to alleviate the differences on QoS requirements in the web services discovery. The proposed approach is implemented and evaluated through an example of hotel booking web service.

1. Introduction Web services have become a promising technology for e-trading and the development of information systems. However, the web service providers and consumers not only expect provisions of services with required functional aspects, but also demand good quality of services (QoS) with non-functional aspects such as service reliability, security, trust and execution cost. QoS is therefore considered as important criteria for service evaluation in the service discovery process. Locating required web services by examining QoS is a typical non-functional service discovery approach when there are a number of services providing same or similar functions. The service providers have to advertise the services in an appropriate and concise manner. In addition, the advertisement needs to be in line with service consumers’ expectations. However, due to the dynamic and unpredictable nature of web services, a concrete measurement for depicting appropriate QoS is not an easy task. QoS can be measured using various attributes. For instance, W3C [2] define various QoS attributes such as performance, reliability, scalability, capacity, and so on. Though W3C provides a list of QoS attributes, their evaluation still involves greater complexity, especially, when fuzzy terms (e.g. good quality, very reliable, and good performance) are employed to express service requests and advertisements

in the discovery process. Such definitions of the fuzzy terms vary according to the opinions and preferences of service consumers and service providers. For instance, a service with 90% success rate is considered as reliable by some service consumers and providers, but others may demand 99% reliability. Similarly, a service with 8 ms response time can be considered as efficient by some while others may rate this service as inefficient. Even though the terms or criteria for QoS have been defined explicitly, the service consumers and providers could still have different expectations of these fuzzy terms. Traditional ontological engineering approach assumes that the shared conceptualization is defined by a bounded set of participants. However, in the process of service discovery, the participants can leave and join the group at any point of time. Thus, the definitions of the terms can evolve over the time in order to meet the participants’ demands. Therefore, the ontology with static nature cannot meet the requirements derived from dynamic environment. This paper proposes a consensus-based QoS moderation method, called QCMA (QoS Consensus Moderation Approach) which is built on MFDM (Moderated Fuzzy Discovery Method) framework [1] and RMGDP (Resolution Method for Group Decision Problems) [3][4][5][6][7]. It attempts to provide a kind of architecture for a group to reach consensus on the QoS. It assumes that a set of QoS characteristics have been defined. These are based on the QoS characteristics defined by W3C [2]. A group of service providers and consumers can reach a consistent consensus on the definition of QoS. As a result, their preferences and expectations can be moderated in order to have coherent fuzzy QoS terms. The paper is organized as follows. Section 2 describes MFDM and RMGDP methods which are the key components of the proposed QCMA framework. Section 3 presents QCMA framework. Section 4 reports on experimental results with a simple case study. Finally, conclusion and the future work are illustrated in Section 5.

2. Moderated Fuzzy Discovery Method (MFDM) MFDM [1] was proposed to achieve an effective web service discovery through moderated fuzzy matchmaking. It not only measures the similarity between the services in

Proceedings of the 20th International Conference on Advanced Information Networking and Applications (AINA’06) 1550-445X/06 $20.00 © 2006

IEEE

terms of capability, syntax and semantics [8][9], but also involves the use of underlying data and information as search criteria. MFDM was built upon fuzzy logic, semantic web, and decision support methods to form a framework along with a set of procedures for service consumers and providers so that they can reach consensus on the representation of contents of services [10]. A built-in domain dependent fuzzy classifier is employed to classify large amount of data and information stored in repositories into concise semantics representation for service discovery. The moderation process is initiated by a fuzzy moderator to minimize the differences among service consumers and providers. The feedback from consumers for vague queries can be tracked and help to categorize more similar terms into fuzzy classes. In order to carry out the “consensus-based” web service discovery by using the proposed fuzzy moderator, the MFDM consists of two parts: Functional Deployment in MFDM Framework and Similarity Aggregation Method.

2.1 Functions Deployment in MFDM Framework MFDM framework consists of a number of components including Fuzzy Classifier, Fuzzy Engine, UDDI / OWL and Fuzzy Moderator. The fuzzy classifier including built-in domain knowledge is able to interpret raw data stored in the service provider’s repositories and represent them with fuzzy terms. These fuzzy terms will be employed by the service provider to advertise their service via UDDI. Since UDDI does not have facility for modeling semantic, the OWL is used for capturing the semantics. The opinions and preferences given by the service providers and consumers will be processed via fuzzy moderator in order to identify their consensus. This enables service consumers (issuing vague requests) and the service providers (using different terms for service advertisement) to moderate their expectations. The following sub-sections describe the Similarity Aggregation Method and Resolution Method for Group Decision Problems, which are used in the moderation process. .

2.2 Similarity Aggregation Method SAM was developed for resolving conflicts emerged from different opinions. In SAM different fuzzy opinions will be aggregated into opinion consensus class so that they can be measured by their similarities. Therefore, the similarity measuring method is the key to generate consensus index in the fuzzy opinions set. This characteristic was taken by the Fuzzy Moderator for moderating definitions of fuzzy terms. During the process of fuzzy term moderation, the consensus indexes are collected and a consensus agreement is formed. The

procedure to perform SAM is organized into 7 steps stated below [1]: 1. Each participant represents his/her subjective fuzzy preference on one specific criterion with a positive trapezoidal fuzzy number. 2. This step is to obtain the opinion similarity between any two participants for the specific criterion. 3. To build an agreement matrix for showing each similarity between each pair of participants in the group is the main task. 4. This step is to calculate average agreement degree for each single participant i, denoted A(Useri ) , in the group. 5. Obtaining RAD (Relative Agreement Degree) for each individual participant using the following formula is carried out in this step. RAD (User i ) =

A (User i ) all

∑

A (User j )

6. It involves individual’s assignment of weighting variable wi to each criterion. 7. To obtain the individual CDC (Consensus Degree Coefficient) for each participant: (2) CDC (Useri ) = β × wi + (1 − β ) × RAD (Useri ) where ȕ is a correlative control variable to indicate the relation between CDC and RAD. If ȕ = 0, then CDC is completely equivalent to RAD.

2.3. Resolution Method for Group Decision Problems (RMGDP) Opinion similarity partially enables the service consumers and providers to reach consensus on QoS of web services, since they may have different preferences on the criteria. Therefore, RMGDP is proposed to alleviate their differences on preferences. RMGDP includes the following three steps:

2.3.1. The transformation phase In the transformation phase, all participants will be grouped. Each participant has to evaluate alternatives according to given criteria, and to assign preference orders to the related alternatives. The participants allocate the orders based on their preferences and subjective judgments. A transfer function, p ijk , is defined for converting these orders of alternatives to a preference relation which sets the ordering preference degree between alternative ai and a j , denote as oi and o j . This is can be expressed by participant Userk as follows: p ijk = f (o ki ,o kj ) =

o kj ok 1 (1 + − i ) 2 m −1 m −1


IEEE

(1)

j −1

(3)

where pijk is a preference relation for Userk by subjective preference order for the ai over a j . m indicates the number of alternatives in the analysis.

2.3.2. The aggregation phase In the aggregation phase, pijc is defined for aggregating participants’ preferences order { pij1 ,..., pijn } by means of fuzzy majority [12]. The fuzzy majority is formed with the OWA (Ordered Weighted Averaging) operator FQ with the

where d cji = max{ p cji − p ijc ,0} . According to (4) the (6) can be represented as: (7) FQ (1− d 1ji ,1− d 2ji ,...,1− d nji ) = w1 .(1− b1 ) + ... + wn .(1− bn )

(

)

where bi is the i-th small value in d 1ji , d 2ji , , d nji The solution offered by Eq. (6)(7) indicates that the fuzzy majority in remaining alternatives a j ( j =1,...,m) cannot dominate the alternative a i . Also, all the preferences in the alternatives can be prioritized and the corresponded order can be obtained.

fuzzy quantifier Q. The function with FQ and Q aggregates

(b). Quantifier guided Dominance Degree

the individual preference values for obtaining the preference order from n users via the following formula:

QGNDD cannot be used for ordering among the preferences if the obtained u NDD ( a i ) from numerous alternatives is in the situation of Unfuzzy Nondominated (UND) [13], i.e., u NDD ( a i ) = 1 . For instance, if

n

pijc = FQ ( pij1 ,..., pijn ) = ∑ wi bi i =1

where bi is the i-th large value in

(4)

(p

1 ij

, p ij2 , , p ijn

)

and wi = Q(i / n) − Q(i − 1 / n) .

2.3.3. The exploitation phase In the exploitation phase, the exploitation process is to obtain the consequence of identifying each alternative priority in group preference. In this process, two well-known fuzzy ranking methods: Quantifier guided Non-Dominance Degree (QGNDD) and Quantifier guided Dominance Degree (QGDD) [13] are taken for alternative priority evaluation.

u NDD ( a i ) ≥ 0 . 8 which is calculated by (5) indicates the “most” quantifier, then UND situation occurs. Also, in order to avoid more than 2 UND situations occurring simultaneously, the obtained fuzzy preference orders need to be validated by other fuzzy ranking methods such as QGDD. By [3], QGDD can quantify the dominance for each ai which has preference order over all other alternatives. QGDD can be taken for prioritizing the final ordering preference with the equation defined as below: m

QGDD (a i ) = FQ ( p ijc , j = 1...m, i ≠ j ) = ∑ wi bi

(8)

i =1

(a). Quantifier guided Non-Dominance Degree (QGNDD) QGNDD is a fuzzy ranking method by means of fuzzy preference relations. The method determines the relative preference degree of alternatives. The Non-Dominance Degree (NDD) of fuzzy ranking can be calculated by participants’ preference relation, which is formulated as follows: (5) u NDD = 1 − max{ p cji − pijc ,0} From Eq. (5), the membership function u NDD ( a i ) can be interpreted as the degree to which ai is not dominated by any others a j ( j = 1,..., m ) . The function u NDD ( ai ) is taken for finding the highest order of alternatives. The NDD for alternative ai is taken for quantifying a criterion which has a higher preference degree than the others. For a linguistic quantifier Q (e.g. “most”), the NDD of the linguistic quantifier is represented as Quantifier Guided Non-Dominance Degree (QGNDD) defined as below: (6) QGNDD(a i ) = FQ (1 − d cji , j = 1...m, j ≠ i)

where wi = Q (i / m ) − Q (i − 1 / m) . By (8) the UND situation can be solved and final preference ranking for a1 , a2 ,...,am can be determined.

3. The Framework and Operation for QCMA Existing research defines various QoS characteristics for web services such as reliability, performance, integrity, etc. Even if a set of standardized characteristics of QoS requirements for web services can be accepted by the community, the opinions of these characteristics can still be inconsistent. In addition, QoS should be treated as a set of dynamic variables rather than a fixed set due to frequent change of their functionalities through service composition. So, the proposed framework will focus on the design of a framework that can facilitate reaching consensus on the opinions of Web service QoS. Since this research is more interested in opinions of QoS than the characteristics of QoS, the QoS requirement defined by W3C is adopted for modeling the domain knowledge which is implemented in the QCMA.. The QoS opinion set is defined as S QoS which comprises the QoS terms shown as below:


IEEE

S QoS = {a1 , a 2 , a3 , , a13 }

(9)

where a1 = “Performance”, a 2 = “Reliability”, a 3 = “Scalability”, a 4 = “Capacity”, a 5 = “Robustness”, a 6 = “Exception Handling”, a 7 “Accuracy”, a8 = “Integrity”,

a9 = “Accessibility”,

a10 = “Availability”,

a11 =

“Interoperability”, a12 = “Security” and a13 = “Network-Related QoS Requirement”. Additional QoS aspects can be introduced into S QoS , the same process will be applied to these aspects in the proposed QCMA. The framework of QCMA is shown as below: QoS Feedback

Services Discovered

Services Feedback

Vague Request

Services Description

Vague Request

Rules Analysis Services Discovered

Rules Analysis

Service Information

Services in Fuzzy Values

Fuzzy Discovery

QoS Feedback

Services Registry

UDDI OWL-S

End User

Fuzzy Classifier

Fuzzy Engine

Service Information

End User

Inference Rules

Services Provisioning

OWL

Rules Analysis

Fuzzy Moderator

Services Feedback

Semantic Analysis

Semantic Analysis

User Preference

Service Information

Semantic Analysis Fuzzy QoS Moderator

interactions between service consumers and providers, called Web Service Activity (wsa), which must be processed by SAM and RMGDP. Through SAM, a threshold value for CDC, t cdc , is set for clustering all input wsa into corresponded S wsa (x ) defined as below: (10) S wsa ( x) = {x is index for wsa | CDC(wsa) ≥ t cdc } After the SAM process is completed, each wsa in

S wsa (x ) will be further processed for preference analysis via RMGDP. The preference order of QoS terms for each wsa will be ranked. Since wsa comprises the fuzzy web service terms given by service consumers and providers, the result generated by SAM and RMGDP will be treated as the input for analyzing the QoS consensus correlation between wsa and QoS terms in S QoS . In the process of web service discovery, the opinions for each current consensus QoS term are updated accordingly. The process includes opinion similarity calculation and preference analysis for each pair of QoS terms. Suppose a web service class k derived from k S wsa ( x ) is generated, the corresponding QoS Opinion Set, k k denotes as S QoS , is produced in association with S wsa ( x) . k The QoS Opinion Function, FQoS defined in the Fuzzy QoS Moderator, starts to reason over wsa via the function shown as below: ~ ~ ~ k (11) FQoS ( wsa) = ( E ak ( wsa), Eak ( wsa),, Eak ( wsa)) 1

n

2

where each non-linear fuzzy evaluation function ~k E a ( wsa ) is represented as below: i

Rules Analysis QoS Discovered

Consensus QoS Base

Figure 1: The Framework of QCMA QCMA includes two key mechanisms. One mechanism includes Fuzzy Classifier, UDDI / OWL-S, Fuzzy Engine, Fuzzy Discovery and Fuzzy Moderator whose functions are the same as or similar to those in MFDM. The other mechanism contains Fuzzy QoS moderator and Consensus-based QoS which are designed for performing QoS consensus. The consensus QoS based comprises group consensus opinions and preferences of the terms (e.g. good, fair, bad etc) for QoS characteristics in S QoS . The opinions for QoS terms are updated regularly by the fuzzy QoS moderator in line with the interactions taking place between the participants. The QoS and their moderated opinions are represented in OWL as ontology. However, the participants have their private spaces to store their opinions. A mechanism is devised for mapping their subjective opinions onto the moderated opinions, when a request is initiated. Each action for QoS consensus moderation is driven by the

⎧ ⎪ l1 ⎪ ~k ⎪⎪ l E ai ( wsa ) = ⎨ 2 ⎪ ⎪ ⎪l p ⎪⎩

1 0 ≤ μ~ai ( wsa ) < + q1 p 2 1 ~ ≤ μ ai ( wsa ) < + q 2 p p p −1 ~ ≤ μ ai ( wsa ) ≤ 1 p

~ For each Eaki (wsa) the returned value for QoS

terms ai can be scaled from l1 to l p and p is the number of possible terms (e.g. very good, good, fair, poor, very poor) that can be used for expressing the opinions. The returned QoS opinion will be determined by the fuzzy membership function μ~a ( wsa ) . qi (i = 1~p-1) is a control i

variable for E~ak ( wsa ) which represents a fuzzy QoS scale. i

Therefore, through the analysis of the fuzzy correlation k , the wsa can be reasoned in terms of similarity and FQoS k group preference among QoS terms in S QoS so that the

QoS consensus moderation can be executed. An experiment for explaining QCMA is given in following section.


IEEE

(12)

4. Experiment: Hotel Booking Web Service

⎡ . 50 ⎢ . 29 ⎢ ⎢ . 63 ⎢ ⎢ . 54 ⎢ . 33 ⎢ ⎢ . 38 1 p = ⎢ . 58 ⎢ ⎢ .67 ⎢ . 46 ⎢ ⎢ .17 ⎢ ⎢ . 42 ⎢ . 25 ⎢ ⎣ . 21

A number of hotel booking web services are adopted in order to illustrate the effectiveness of the proposed system. Assume that a hotel booking is denoted as a web service k activity wsa hbs . It includes a set of QoS terms, S QoS that can be used by tourist k, which are defined as below: k (13) wsa hbs = ( a~1k , a~2k , , a~13k ) k ~k k ~ ~ where ( a1 , a 2 , , a13 ) follows the definition in S QoS . In this experiment, 4 tourists were invited to conduct a group QoS consensus analysis, which involves 4 web services using the proposed QCMA. The opinions about k web service activities wsa hbs from the 4 tourists were k k ~ ~ transformed as ( a , a , …, a~ k ) which are positive 1

2

⎡. 50 ⎢ .33 ⎢ ⎢ .75 ⎢ ⎢. 67 ⎢. 58 ⎢ ⎢. 46 p 2 = ⎢ .63 ⎢ ⎢ . 71 ⎢. 42 ⎢ ⎢ .25 ⎢. 54 ⎢ ⎢. 29 ⎢. 38 ⎣

13

k trapezoidal fuzzy numbers and form a wsa hbs matrix shown as below:

wsa

wsa

wsa

wsa

1 hbs

2 hbs

3 hbs

4 hbs

⎡0 ⎢0 =⎢ ⎢7 ⎢ ⎣8 ⎡0 ⎢0 = ⎢ ⎢6 ⎢ ⎣8 ⎡0 ⎢0 = ⎢ ⎢8 ⎢ ⎣9

0 0

⎡0 ⎢0 = ⎢ ⎢3 ⎢ ⎣8

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

5

4

7

4

5

7

3

6

9

6

4

9

7

8

8

8

8

8

9

10

7

8

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

6

5

7

3

4

5

6

5

6

7

7

9

9

8

7

9

10

9

0 0

0 0

9 0 0

5 0 0

7 0 0

9 0 0

0 0

0 0

0 0

0 0

0 0

6

3

5

6

4

7

5

5

6

3

5

10

8

7

9

8

9

8

7

9

7

7

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

4

6

7

5

6

7

4

7

5

3

6

7

8

8

7

9

9

7

10

9

8

9

0⎤ 0⎥ ⎥ 6⎥ ⎥ 9⎦ 0⎤ 0⎥ ⎥ 4⎥ ⎥ 7⎦ 0⎤ 0⎥ ⎥ 6⎥ ⎥ 8⎦ 0⎤ 0⎥ ⎥ 4⎥ ⎥ 7⎦

⎡.50 ⎢. 75 ⎢ ⎢. 88 ⎢ ⎢.96 ⎢. 83 ⎢ ⎢ .71 3 p = ⎢1 . 0 ⎢ ⎢.92 ⎢.67 ⎢ ⎢. 58 ⎢ ⎢.79 ⎢.54 ⎢ ⎣. 63 ⎡ .50 ⎢ .42 ⎢ ⎢ .58 ⎢ ⎢ .71 ⎢ . 33 ⎢ ⎢ .29 p 4 = ⎢. 67 ⎢ ⎢ . 63 ⎢ .46 ⎢ ⎢ .25 ⎢ ⎢ .54 ⎢ .38 ⎢ ⎣ .21

Using SAM, the average agreement degree for 4 k can be obtained as follows: wsa hbs wsa

avr hbs

⎡0 ⎢0 =⎢ ⎢6 ⎢ ⎣8

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

0 0

5

5

9

8

7

5

5

7

5

6

7

5

6

8

7

8

9

8

8

9

8

8

0⎤ 0⎥ ⎥ 5⎥ ⎥ 8⎦

k With the use of CDC for 4 wsa hbs (k = 1 ~ 4), all the opinions were treated as “being accepted and classified into a consensus in term of opinion similarity”. The RMGDP is employed to identify the possible compromised preference orders from their diverse k preferences. Assume the order preference for the 4 wsa hbs 2 3 4 , ohbs , ohbs ) with (k = 1 ~ 4) is denoted as ( o1hbs , ohbs corresponding preference orders shown as below: o1hbs = {a8 , a3 , a7 , a 4 , a1 , a 9 , a11 , a 6 , a5 , a 2 , a12 , a13 , a10 } 2 ohbs = {a3 , a8 , a 4 , a7 , a11 , a 5 , a1 , a 6 , a9 , a12 , a 2 , a13 , a10 } 3 ohbs = {a7 , a 4 , a8 , a 3 , a5 , a11 , a 2 , a 6 , a9 , a13 , a10 , a12 , a1 }

4 ohbs = {a 4 , a7 , a8 , a 3 , a11 , a1 , a9 , a 2 , a12 , a 5 , a 6 , a10 , a13 }

2 3 4 , o hbs , o hbs ), the ( p1 , p 2 , According to ( o1hbs , o hbs p 3 , p 4 ) can be obtained via RMGDP as below:

. 71

. 38

. 46

. 67

. 63

. 42

. 33

. 54

.83

.58

.75

. 50 . 83

. 17 . 50

.25 .58

.46 .79

.42 . 75

. 21 . 54

. 13 . 46

. 33 . 67

.63 . 96

.38 . 71

. 54 .88

. 75

. 42

. 50

.71

. 67

. 46

. 38

. 58

. 88

.63

. 79

. 54 . 58

. 21 . 25

. 29 . 33

.50 .54

.46 .50

. 25 . 29

.17 . 21

. 38 . 42

. 67 . 71

. 42 . 46

.58 . 63

. 79 . 88

. 46 . 54

. 54 . 63

. 75 . 83

.71 .79

. 50 . 58

. 42 . 50

. 63 . 71

. 92 1.0

. 67 .75

. 83 . 92

. 67 . 38

.33 . 04

. 42 . 13

. 63 . 33

. 58 .29

. 38 . 08

. 29 . 00

. 50 . 21

. 79 . 50

. 54 .25

. 71 . 42

. 63 . 46

. 29 .13

.38 . 21

. 58 .42

.54 . 38

. 33 .17

. 25 . 08

. 46 . 29

. 75 . 58

. 50 .33

. 67 . 50

. 42

. 08

. 17

. 38

. 33

. 13

. 04

. 25

. 54

. 29

. 46

.67

.25

.33

.42

. 54

. 38

.29

. 58

. 75

.46

. 71

.50

.08

. 17

. 25

.38

.21

. 13

. 42

. 58

. 29

. 54

.92 . 83

. 50 . 42

. 58 . 50

.67 .58

. 79 . 71

.63 . 54

.54 .46

.83 .75

1 .0 .92

. 71 . 63

. 96 . 88

. 75 . 63

. 33 . 21

. 42 . 29

.50 .38

.63 . 50

. 46 .33

.38 . 25

. 67 . 54

. 83 .71

. 54 . 42

. 79 . 67

.79

.38

. 46

.54

. 67

. 50

.42

. 71

. 88

.58

.83

. 88 . 58

. 46 . 17

. 54 . 25

. 63 . 33

.75 . 46

. 58 . 29

.50 . 21

. 79 . 50

. 96 .67

. 67 .38

. 92 .63

.42

. 00

. 08

.17

. 29

.13

.04

.33

. 50

. 21

. 46

.71 .46

. 29 . 04

. 38 .13

.46 .21

.58 .33

. 42 . 17

. 33 .08

.63 . 38

.79 .54

.50 . 25

.75 . 50

.54

. 13

. 21

.29

. 42

. 25

.17

. 46

. 63

. 33

. 58

. 25

. 13

.04

.17

.29

.00

.08

. 33

.42

.21

.46

.50 . 63

. 38 .50

.29 .42

.42 .54

.54 . 67

.25 .38

. 33 .46

.58 .71

. 67 .79

.46 .58

.71 . 83

.71

. 58

.50

. 63

. 75

.46

.54

.79

.88

. 67

.92

. 58 .46

.46 . 33

. 38 . 25

.50 . 38

. 63 .50

. 33 .21

.42 .29

. 67 .54

.75 . 63

.54 .42

.79 . 67

. 75 .67

. 63 .54

.54 .46

.67 . 58

.79 .71

.50 .42

.58 .50

. 83 .75

.92 . 83

.71 . 63

.96 .88

.42 . 33

.29 .21

.21 . 13

. 33 . 25

.46 . 38

. 17 .08

.25 . 17

.50 .42

.58 .50

.38 .29

. 63 .54

.54 .29

.42 .17

. 33 . 08

.46 .21

. 58 . 33

.29 .04

.38 . 13

. 63 .38

.71 .46

.50 .25

.75 .50

. 38

. 25

.17

.29

.42

. 13

.21

.46

.54

. 33

.58

. 58

.42

. 29

.67

. 71

. 33

.38

. 54

.75

. 46

. 63

. 50 . 67

. 33 .50

. 21 . 38

. 58 . 75

.63 . 79

. 25 .42

. 29 . 46

. 46 . 63

. 67 . 83

. 38 . 54

.54 .71

. 79

. 63

. 50

. 88

. 92

.54

.58

. 75

.96

. 67

. 83

. 42 . 38

. 25 .21

.13 . 08

.50 .46

. 54 . 50

.17 . 13

. 21 . 17

. 38 . 33

.58 .54

. 29 . 25

.46 .42

. 75 . 71

. 58 .54

. 46 . 42

. 83 .79

.88 .83

.50 .46

. 54 . 50

. 71 .67

.92 .88

. 63 . 58

.79 . 75

. 54 . 33

. 38 . 17

. 25 . 04

. 63 .42

. 67 . 46

.29 . 08

.33 .13

. 50 . 29

.71 .50

. 42 . 21

. 58 . 38

. 63 . 46

.46 .29

.33 . 17

.71 .54

.75 .58

. 38 .21

. 42 .25

. 58 . 42

.79 . 63

. 50 . 33

. 67 .50

. 13

. 00

. 38

. 42

.04

.08

. 25

.46

. 17

. 33

. 29

With wiinit

IEEE

. 63 ⎤ .46 ⎥ ⎥ . 88 ⎥ ⎥ .79 ⎥ .71 ⎥ ⎥ . 58 ⎥ . 75 ⎥ ⎥ . 83 ⎥ .54 ⎥ ⎥ . 38 ⎥ .67 ⎥ ⎥ .42 ⎥ .50 ⎥⎦ .38 ⎤ . 63 ⎥⎥ .75 ⎥ ⎥ . 83 ⎥ .71 ⎥ ⎥ .58 ⎥ .88 ⎥ ⎥ .79 ⎥ .54 ⎥ ⎥ .46 ⎥ ⎥ . 67 ⎥ .42 ⎥ ⎥ .50 ⎦ . 79 ⎤ . 71 ⎥⎥ . 88 ⎥ ⎥ 1 .0 ⎥ .63 ⎥ ⎥ . 58 ⎥ . 96 ⎥ ⎥ . 92 ⎥ . 75 ⎥ ⎥ . 54 ⎥ ⎥ .83 ⎥ . 67 ⎥ ⎥ . 50 ⎦

the

initial weight vector = (.25 .25 .25 .25 ) (i = 1 ...4) the collective

preference p c can be: ⎡.50 ⎢. 45 ⎢ ⎢.71 ⎢ ⎢.72 ⎢.52 ⎢ ⎢.46 p c = ⎢.72 ⎢ ⎢. 73 ⎢.50 ⎢ ⎢.31 ⎢ ⎢.57 ⎢.36 ⎢ ⎣. 35

.55 . 50

.29 .24

. 28 . 23

. 48 . 43

. 54 . 49

.28 .23

. 27 .22

. 50 . 45

. 69 . 64

. 43 .38

.64 .58

. 76 . 77

.50 .51

.49 .50

.69 .70

. 75 . 76

. 49 . 50

. 48 .49

. 71 . 90 . 72 .91

.64 .65

.84 .85

. 57 . 51 . 77

.31 .25 .51

.30 .24 .50

.50 .44 .70

. 56 . 50 . 76

. 30 . 24 . 50

.29 . 23 .49

. 52 . 46 . 72

.71 .65 .91

.45 .39 .65

.66 .59 .85

.78 .55

.52 .29

.51 . 28

.71 . 48

. 77 . 54

. 51 .50 .28 . 27

.73 . 50

. 92 . 69

.66 . 43

.86 .64

. 36 .63

.10 .36

.09 . 35

.29 . 55

. 35 . 09 . 61 .35

. 08 .34

. 31 . 50 . 57 . 76

.24 .50

.45 .71

. 42 . 41

.16 .15

. 15 .14

.34 . 33

. 41 .15 . 40 . 14

.14 . 13

. 36 . 35

.29 .28

.50 .49


.79 ⎤ . 58 ⎥⎥ .92 ⎥ ⎥ . 83 ⎥ . 63 ⎥ ⎥ . 67 ⎥ . 88 ⎥ ⎥ .96 ⎥ . 75 ⎥ ⎥ .46 ⎥ ⎥ .71 ⎥ .54 ⎥ ⎥ .50 ⎦

.55 . 54

. 65 ⎤ . 59 ⎥⎥ . 85 ⎥ ⎥ . 86 ⎥ .67 ⎥ ⎥ . 60 ⎥ . 86 ⎥ ⎥ . 88 ⎥ . 65 ⎥ ⎥ . 46 ⎥ ⎥ . 72 ⎥ . 51 ⎥ ⎥ . 50 ⎦

As a result, a1 and a 9 have the same preference order. In addition, a 4 and a 7 have the same preference order. According to sequence of index, let the preference order of a1 be higher than a 9 and let the preference order of a 4 be higher than a 7 . Therefore, the preference c will be: order for QoS consensus in o hbs

c ohbs = {a8 , a4 , a7 , a3 , a11, a5 , a1, a9 , a6 , a2 , a12 , a13, a10}

and the group consensus weighting via GNDD can be:

c wQGDD = (.072 .065 .103 .104 .075 .066 .104 .106 .072 .045 .083 .053 .051)

Also, according to p c , the QoS scale in E~ak (wsa) can be set as 3 and q1 = q 2 = 1 so that each responded wsa from service consumers and providers can be analyzed by QCMA, clustered into right S wsa ( x ) , ranged into i

k appropriate fuzzy scale via FQoS to achieve group QoS consensus. So, the service providers can redefine their opinions in the advertisement and the service consumers are able to use the mapping mechanism to retrieve their required services.

6. Conclusion The work carried out in this research focused on the QoS-based web services discovery. It proposes a QoS Consensus Moderation Approach (QCMA) in order to perform QoS consensus and to alleviate the differences on QoS requirements in the web services discovery.ʳ The proposed QCMA possesses the following features. 1. QCMA is a web service discovery mechanism based on fuzzy QoS consensus for a group of participants. The architecture allows them to reach QoS consensus by including a number of activities such as participants’ opinion similarity, QoS term preference ordering and QoS fuzzy scale for each QoS term. By the contribution QCMA will not only be taken for recognizing the inquiry of web service discovery in fuzzy way, but also offering the features to predict the QoS preference consensus after aggregating sufficient wsa. 2. QCMA is designed for open and dynamic web environment, so new opinions and preferences as well as new QoS aspects can be modeled flexibly. 3. The functionality of QoS Opinion Function in QCMA can be executed in parallel. This potentially can increase the efficiency. The future work of this research will focus on the investigation of other intelligent approaches such as neural network, genetic algorithm, etc. to improve weighting determination for alternatives preference orders in QCMA.

Also, for preventing too big error between individual opinion and QoS consensus, the multi-QoS consensus in distributive communities is deserved to study.

References [1] C-L Huang, K-M Chao, C-C Lo: A Moderated Fuzzy Matchmaking for Web Services, (to appear in) Proceedings of The 5th International Conference on Computer and Information Technology, Shanghai, China, IEEE CS, 2005 [2] W3C, QoS for Web Service: Requirements and Possible Approaches, W3C Working Group Note 25 November, http://www.w3c.or.kr/kr-office/TR/2003, 2003 [3] E Herrera-Viedma, F. Herrera, and F. Chiclana: A Consensus Model for Multiperson Decision Making With Different Preference Structures. IEEE Transactions on Systems, Man and Cybernetics, IEEE, Vol.32.394–402, 2002 [4] F. Chiclana, F Herrera, and E. Herrera-Viedma: Integrating Multiplicative Preference Relations in A Multipurpose Decision-making Model Based on Fuzzy Preference Relations. Fuzzy Sets and Systems, Elsevier Science, Vol. 122. 277–291, 2001 [5] F. Chiclana, F. Herrera, E. Herrera-Viedma: A Classification Method of Alternatives for Multiple Preference Ordering Criteria Based on Fuzzy Majority. Journal of Fuzzy Math., Vol. 34, 224–229, 1996 [6] F. Chiclana, F. Herrera, E. Herrera-Viedma: Integrating Three Representation Models in Fuzzy Multipurpose Decision Making Based on Fuzzy Preference Relations. Fuzzy Sets and Systems, Elsevier Science, Vol. 97. 33–48, 1998 [7] F.Herrera et al.: A Rational Consensus Model in Group Decision Making Using Linguistic Assessments. Fuzzy set and Systems, Elsevier Science, Vol. 88, 31–49, 1997 [8] K. Sycara, S. Widoff: Dynamic Matchmaking Among Heterogeneous Software Agents in Cyberspace, Journal of AAMAS, 5, Kluwer Academic Publishers, 2002. [9] Y. Wang, E. Stroulia: Flexible Interface Matching for Web-Service Discovery, Proceedings of Fourth International Conference on Web Information System Engineering, IEEE CS, December, 2003 [10] Kuo-Ming Chao, Muhammad Younas, Chi-Chun Lo, Tao-Hsin Tan: Fuzzy Matchmaking for Web Service, Proceedings of 19 IEEE Conference on Advanced Network and Information Application, IEEE CS, pp 721-726, 2005 [11] R.R. Yager: On Ordered Weighted Averaging Aggregation Operators in Multi Criteria Decision Making. IEEE Trans. on Systems, Man and Cybernetics, IEEE, Vol. 18. 183-190, 1988 [12] F. Chiclana, F. Herrera, E. Herrera-Viedma: A Classification Method of Alternatives for Multiple Preference Ordering Criteria Based on Fuzzy Majority. Journal of Fuzzy Math., Vol. 34, 224–229, 1996 [13] S.A. Orlovski: Decision-Making with A Fuzzy Preference Relation. Fuzzy Sets and Systems, Elsevier Science, Vol.1. 155-167, 19c


IEEE