Sie sind auf Seite 1von 12

Available online at www.sciencedirect.

com

Information Sciences 178 (2008) 10981109 www.elsevier.com/locate/ins

A method for group decision making with multi-granularity linguistic assessment information
Yan-Ping Jiang
a

a,*

, Zhi-Ping Fan a, Jian Ma

Department of Management Science and Engineering, School of Business Administration, Northeastern University, Shenyang 110004, China b Department of Information Systems, City University of Hong Kong, Kowloon, Hong Kong, China Received 26 December 2005; received in revised form 5 September 2007; accepted 11 September 2007

Abstract This paper proposes a method to solve the group decision making (GDM) problems with multi-granularity linguistic assessment information. In the method, the multi-granularity linguistic information provided by experts is rstly expressed in the form of fuzzy numbers. In order to make the collective opinion close to each experts opinion, a linear goal programming model is constructed to integrate the fuzzy assessment information and to directly compute the collective ranking values of alternatives without the need of information transformation. Then, a fuzzy preference relation on the pairwise comparisons of the collective ranking values of alternatives is constructed using the dominance possibility degree of the comparison between the fuzzy numbers. By applying a non-dominance choice degree to this fuzzy preference relation, the ranking of alternatives is determined and the most desirable alternative(s) is selected. An example is used to illustrate the applicability of the proposed method and its advantages. 2007 Elsevier Inc. All rights reserved.
Keywords: Group decision making (GDM); Linguistic assessment information; Multi-granularity; Fuzzy number; Goal programming; Alternative ranking

1. Introduction Group decision making (GDM) problems with linguistic information arise from a wide range of real-world situations [6,15,16]. In linguistic GDM analysis, rstly, experts provide their assessment information from the pre-established linguistic term sets. Then the linguistic information provided by experts is aggregated to form a collective opinion on the alternatives and the most desirable alternative(s) can be selected according to the derived collective opinion [7,13,19]. Most of the proposals for solving GDM problems with linguistic information have been found in literature are focused on the cases where the information provided by experts is represented in the same linguistic term set [11,14]. However, in practical GDM problems, the experts have their dierent cultural, educational backgrounds, experience and knowledge. Also their decisions can be made under dierent circumstances.
*

Corresponding author. Tel.: +86 24 23891100; fax: +86 24 23891569. E-mail address: ypjiang@mail.neu.edu.cn (Y.-P. Jiang).

0020-0255/$ - see front matter 2007 Elsevier Inc. All rights reserved. doi:10.1016/j.ins.2007.09.007

Y.-P. Jiang et al. / Information Sciences 178 (2008) 10981109

1099

The experts maybe use linguistic term sets with dierent cardinalities, i.e., multi-granularity linguistic term sets, to express their individual assessment information [10]. This type of information is referred to as multi-granularity linguistic information [10,12]. For instance, in a GDM problem of selecting R&D projects, some experts are willing to use a linguistic term set with two terms (e.g., 1: Pass and 2: Fail) while others prefer to use the one with ve terms (e.g., 1: Excellent, 2: Very Good, 3: Good, 4: Pass and 5: Fail). It is necessary to point out that the linguistic term set with small cardinality is benecial for the experts to express their clear assessment information, while the linguistic term set with large cardinality provides experts with more choices to express their exact assessment information. Therefore, the research on GDM problems with multi-granularity linguistic information is important to real applications and several methods have been proposed to solve this kind of GDM problems [10,12]. Herrera et al. [10] investigated a fusion method based on the linguistic 2-tuple representation model to handle the multi-granularity linguistic information. In their method, rstly, the one with maximum granularity in all of the pre-established linguistic term sets was selected as the basic linguistic term set (BLTS). Then, the multi-granularity linguistic information given by experts was converted into the fuzzy sets dened in the BLTS by means of a transformation function. To obtain the collective ranking values of alternatives, these fuzzy sets were aggregated by means of the OWA aggregation operator. Subsequently, the most desirable alternative(s) was selected by using the non-dominance choice degree of alternatives. Herrera and Martinez [12] proposed another method to solve the GDM problem with multi-granularity linguistic information. They constructed linguistic hierarchy term sets and generalized transformation functions to unify the multi-granularity linguistic information into the linguistic 2-tuples. A fuzzy linguistic 2-tuple aggregation operator was used to integrate the unied information into the collective opinion and to compute the collective ranking values of alternatives. The selection of the most desirable alternative(s) was done based on the collective ranking values. Prior studies [10,12] have signicantly advanced GDM analysis with multi-granularity linguistic information. However, unifying the multi-granularity linguistic information increases the steps of computation and the experts subjectivity in determining a transformation function. Current methods also neglect to consider how to narrow the gap between the collective opinion and each experts opinion so as to make the collective opinion reach better group consensus. This paper presents a new method to deal with the GDM problem. The motivation of presenting the method is based on the following facts: (1) The proposed method does not need to unify multigranularity linguistic information. In this paper, the experts multi-granularity linguistic information is integrated and collective ranking values of alternatives are directly obtained through the linear goal programming model. Thus, the proposed method is simpler than the existing ones and can avoid the subjectivity in determining a transformation function. (2) The method makes use of the optimization model. It keeps the collective opinion as close to each experts opinion as possible, thus makes the nal collective opinion reect every experts opinion and improves the group consensus. (3) The method is applicable to multi-granularity linguistic information with both symmetrical membership function and unsymmetrical membership function. The rest of the paper is arranged as follows. In Section 2, the GDM problem with multi-granularity linguistic information is presented. In Section 3, a goal programming model is proposed to integrate the multi-granularity linguistic information and to obtain the collective ranking values of alternatives. The ranking of alternatives is determined and the most desirable alternative(s) is selected by constructing a fuzzy preference relation on the pairwise comparisons of the collective ranking values of alternatives and applying a non-dominance choice degree to this fuzzy preference relation. In Section 4, an example is used to illustrate the proposed method. The last section concludes this paper. 2. Presentation of the problem This section describes the GDM problem with multi-granularity linguistic information. q q Let S q fsq 0 ; s1 ; . . . ; sT q g be the qth pre-established nite and totally ordered linguistic term set with odd q cardinalities, q 1; . . . ; p, where sq h denotes the hth linguistic term of set S . It is seen that T q 1 is the carq q dinality of S . And it is usually required that set S has the following characteristics [2,8]:
q q q (1) The set is ordered: sq t \ P " sh if t P h, where P means that st is preferred or indierent to sh . q q (2) There is a negation operator: negsh st such that t T q h.

1100

Y.-P. Jiang et al. / Information Sciences 178 (2008) 10981109

q q q q (3) There is a maximization operator: maxfsq h ; st g sh if sh \ P " st . q q q q q (4) There is a minimization operator: minfsh ; st g sh if st \ P " sh .

The linguistic term sets, S 1 ; S 2 ; . . . ; S p , are called multi-granularity linguistic term sets. In the considered GDM problem, we assume that dierent experts use the multiple linguistic term sets with distinct granularities to express their individual assessment information. This kind of assessment information is referred to as multi-granularity linguistic information [10]. To facilitate representation and analysis, the following notations are used throughout the paper. Let X fX 1 ; X 2 ; . . . ; X n g be a discrete set of n possible alternatives and D fD1 ; D2 ; . . . ; Dm g be a nite set of m experts, where X i denotes the ith alternative and Dk denotes the kth expert. The expert Dk provides his/ T her assessment information on X as a vector ^ p k ^ pk pk pk 1; ^ 2; . . . ; ^ n according to his/her preferred linguistic term q q k set S , where ^ pi 2 S denotes the assessment value of alternative X i given by expert Dk . Thus, the experts b ^ assessment information can be represented by the following linguistic decision matrix P pk i mn ,

The problem concerned in this paper is how to rank alternatives or select the most desirable alternative(s) among a nite set X based on the multi-granularity linguistic information given by the experts. The resolution process is analyzed in details in the following sections. 3. The proposed method The resolution process of the GDM problem with multi-granularity linguistic information is presented in Fig. 1. In the resolution process outlined above, after the multi-granularity linguistic information is represented in the form of fuzzy numbers, a goal programming model is constructed to assess the collective ranking values of alternatives. Then, the collective ranking values of alternatives in the form of fuzzy numbers are compared to obtain the ranking of alternatives or to select the most desirable alternatives. Based on Fig. 1, the

Fig. 1. Resolution process of the GDM problem.

Y.-P. Jiang et al. / Information Sciences 178 (2008) 10981109

1101

resolution process of the GDM problem with multi-granularity linguistic information is developed as follows. 3.1. Represent the multi-granularity linguistic information in fuzzy numbers Usually, the fuzzy numbers dened in interval [0, 1] are considered to be adequate to capture the uncertainty and vagueness of linguistic information [10]. The fuzzy numbers are described by membership function such as linear trapezoidal membership function [2], linear triangular membership function [1], Gaussian function [4] and so on. In this paper, the trapezoidal fuzzy number is used to represent the linguistic term sq i of set q q q q S q . Its associated representation is ~ sq i ai ; bi ; d i ; ci . As a particular case of this type of representation, a triq q q q q q q angular fuzzy number ~ sq sq i ai ; bi ; ci can be represented as ~ i ai ; bi ; bi ; ci . The multi-granularity linguistic 1 2 3 4 term sets S ; S ; S and S used in this paper are given as follows [10]: S1 s1 0 s1 1 s1 2 s1 3 s1 4 s1 5 s1 6 s1 7 s1 8 S2 s2 0 s2 1 s2 2 s2 3 s2 4 s2 5 s2 6 S3 s3 0 s3 1 s3 2 s3 3 s3 4 S4 s4 0 s4 1 s4 2 s4 3 s4 4 s4 5 s4 6 s4 7 s4 8 Fuzzy number (0, 0, 0.12) (0, 0.12, 0.25) (0.12, 0.25, 0.37) (0.25, 0.37, 0.5) (0.37, 0.5, 0.62) (0.5, 0.62, 0.75) (0.62, 0.75, 0.87) (0.75, 0.87, 1) (0.87, 1, 1) Fuzzy number (0, 0, 0.16) (0, 0.16, 0.33) (0.16, 0.33, 0.5) (0.33, 0.5, 0.66) (0.5, 0.66, 0.83) (0.66, 0.83, 1) (0.83, 1, 1) Fuzzy number (0, 0, 0.25) (0, 0.25, 0.5) (0.25, 0.5, 0.75) (0.5, 0.75, 1) (0.75, 1, 1) Fuzzy number (0, 0, 0, 0) (0, 0.01, 0.02, 0.07) (0.04, 0.10, 0.18, 0.23) (0.17, 0.22, 0.36, 0.42) (0.32, 0.41,0.58, 0.65) (0.58, 0.63, 0.80, 0.86) (0.72, 0.78, 0.92, 0.97) (0.93, 0.98, 0.99, 1) (1, 1, 1, 1)

The linguistic terms in sets, S 1 ; S 2 ; S 3 and S 4 , and their associated fuzzy numbers are graphically shown in Figs. 25, respectively.

1102

Y.-P. Jiang et al. / Information Sciences 178 (2008) 10981109

Fig. 2. Linguistic term set S 1 .

It is seen that the cardinality of linguistic term sets S 1 , S 2 ; S 3 and S 4 are 9, 7, 5 and 9, i.e., T 1 9, 1 2 3 T 2 7; T 3 5 and T 4 9. In particular, the linguistic term, sq i , in sets S ; S and S , can be approximately expressed in the following triangular fuzzy number: q q q ~ sq i 0; . . . ; T q ; q 1; 2; 3; 1 i ai ; bi ; c i ; the membership function of which is 8 q xai q > aq > q q ; i < x < bi ; > < bi ai q ci x q q l~ si x bq q q ; > i < x < ci ; ci bi > > : 0; otherwise;

q q where aq i maxi 1=T q ; 0; bi i=T q and ci mini 1=T q ; 1. According to the associated fuzzy numbers of linguistic terms in multi-granularity linguistic term sets, the b ^ e linguistic decision matrix P pk pk i mn are expressed in the fuzzy decision matrix P ~ i mn , where f k g k k u k v k ~ pi pi ; pi ; pi , pi . In the following, the research problem is how to determine the collective ranking values of alternatives.

Fig. 3. Linguistic term set S 2 .

Fig. 4. Linguistic term set S 3 .

Y.-P. Jiang et al. / Information Sciences 178 (2008) 10981109

1103

Fig. 5. Linguistic term set S 4 .

3.2. Calculate the collective ranking value of each alternative In GDM problem, it is hoped that the nal collective opinion reects every experts opinion [22]. It means the collective ranking values of alternatives are expected to be as close to each experts opinion as possible [3]. Therefore, it is expected to determine the collective ranking value of each alternative so that the deviation degree between each experts opinion and the collective opinion is as small as possible. Suppose the collective f u v ranking value of alternative X i is a fuzzy number ~ p i p g i ; p i ; p i ; p i . In order to measure the deviation degree on alternative X i between the expert Dk s opinion and the collective opinion, the distance d ~ pk pi between the i;~ k fuzzy numbers ~ pi and ~ pi [17] is used, which can be given by d ~ pk pi i;~
f f g u u v v jpg i p i j jp i p i j jp i p i j jp i p i j : 4 k k k k

Based on Eq. (3), the collective deviation degree on alternative X i is given by m m X 1X k f k f g u k u v k v zi d ~ pk pi jpg i p i j jp i p i j jp i p i j jp i p i j: i;~ 4 k 1 k 1 By Eq. (4), the total deviation degree is as follows: n X m n X m X 1X k f k f g u k u v k v ~ z d ~ pk ; p jpg i i p i j jpi p i j jp i pi j jp i pi j: i 4 i 1 k 1 i1 k 1

To make the collective opinion reect each experts opinion, we can minimize z to assess the collective ranking value ~ pi i 1; . . . ; n. Therefore, based on Eq. (5), an optimization model is constructed as follows: n X m X Minimize z d ~ pk p i ; 6a i;~ Subject to Let 1 g k g g k g d ki1 jp i pi j p i p i ; 2 1 g k g g k g d ki1 jp i pi j p i p i ; 2 1 f k f f k f d ki2 jp i p i j p i p i ; 2 1 f k f f k f d ki2 jp i p i j p i p i ; 2 1 u k u u k u d ki3 jp i p i j p i pi ; 2 1 u k u u k u d ki3 jp i p i j p i pi ; 2 1 v k v v k v d ki4 jp i p i j p i p i ; 2 1 v k v v k v d ki4 jp i p i j p i p i ; 2 k 1; . . . ; m; i 1; . . . ; n; k 1; . . . ; m; i 1; . . . ; n; k 1; . . . ; m ; i 1; . . . ; n; k 1; . . . ; m ; i 1; . . . ; n; k 1; . . . ; m; i 1; . . . ; n; k 1; . . . ; m; i 1; . . . ; n; k 1; . . . ; m ; i 1; . . . ; n; k 1; . . . ; m ; i 1; . . . ; n: 7a 7b 7c 7d 7e 7f 7g 7h 06
i1 k 1 f pg i 6 pi v 6 pu i 6 p i 6 1;

i 1; . . . ; n:

6b

1104

Y.-P. Jiang et al. / Information Sciences 178 (2008) 10981109

Then the optimization model (6) is transformed into the following linear goal programming model: Minimize Subject to z
m X n X 4 1X d d kih ; 4 k1 i1 h1 kih k

8a k 1; . . . ; m; i 1; . . . ; n; k 1; . . . ; m ; i 1; . . . ; n; k 1; . . . ; m ; i 1 ; . . . ; n; 8b 8c 8d 8e 8f 8g

g pg i p i d ki1 d ki1 0; d ki2 d ki2 0; pu i d ki3 d ki3 0; pv i d ki4 d ki4 0; f u v 0 6 pg i 6 p i 6 p i 6 p i 6 1; d and d kih P 0; kih P 0 k pf i k pu i k pv i

pf i

k 1; . . . ; m; i 1; . . . ; n;
d kih d kih

k 1; . . . ; m ; i 1 ; . . . ; n; 0; k 1; . . . ; m; i 1; . . . ; n; h 1; 2; 3; 4:

Several methods can be used to solve the problem (8) (please refer to [18,20,21] for details of the calculation f u v process). By solving the goal programming model (8), we can obtain ~ p i p g i ; p i ; p i ; p i ; i 1; . . . ; n. 3.3. Rank the alternatives and chose the most desirable alternatives In order to rank all the alternatives and chose the most desirable alternatives, we should compare the collective ranking values of alternatives. The possibility degree of dominance of one alternative over another one is given below. f u v The collective ranking values of alternatives X i and X j obtained by solving model (8) are ~ p i p g i ; pi ; pi ; pi f g u v and ~ pj pj ; pj ; pj ; pj , respectively. The collective possibility degree of dominance of X i over X j [5,9] is given by pj maxz miny 6z fl~ V ~ pi P ~ pi z; l~ pj y g; 9

where l~ pi and ~ pj , respectively. pi z; l~ pj y are the membership functions of ~ By Eq. (9), a fuzzy preference relation P pij nn on the pairwise comparisons of all the collective ranking values of alternatives can be constructed, where pij V ~ pi P ~ pj for i; j 2 f1; . . . ; ng; i 6 j, and pii for i 1; . . . ; n [10]. For the alternative X i , its non-dominance degree [10], NDDi , can be obtained by NDDi minX j f1 ps ji ; j 6 ig; 10 where ps ji maxfp ji p ij ; 0g represents the degree to which X i is strictly dominated by X j . According to the non-dominance degree, NDD1 ; NDD2 ; . . . ; NDDn , the ranking of alternatives is determined. The greater the value NDDi is, the better the corresponding alternative X i will be. Based on the above analysis, the following algorithm is proposed to solve the GDM problem with multigranularity linguistic information. Algorithm: Step Step Step Step 1 2 3 4 Express the multi-granularity linguistic information in the form of fuzzy numbers. Calculate the collective ranking values of alternatives by model (8). Compute the non-dominance degree of each alternative by Eqs. (9) and (10). Rank the alternatives.

4. Illustrative example In this section, an example about GDM problem from Herrera et al. [10] is used to illustrate the eectiveness of the proposed method. The problem is solved by our method and the fusion method presented in [10], respectively. And then a comparison analysis between the two methods is conducted. Suppose an investment company wants to select the best option to invest a sum of money. There are four possible alternatives to be considered:

Y.-P. Jiang et al. / Information Sciences 178 (2008) 10981109 Table 1 The four experts original linguistic information Expert Alternative X1 D1 D2 D3 D4 s1 4 s2 3 s3 2 s4 4 X2 s1 6 s2 4 s3 3 s4 5 X3 s1 3 s2 3 s3 2 s4 3

1105

X4 s1 5 s2 5 s3 1 s4 5

X 1: X 2: X 3: X 4:

a car industry, a food company, a computer company, and an arms industry.

The investment company has a group of four consultancy departments, i.e., the risk analysis department, the growth analysis department, the socialpolitical analysis department and the environmental impact analysis department. Suppose four experts, D1 ; D2 ; D3 and D4 , from the four dierent departments are invited to participate in the evaluation work. Expertss assessment information is collected through questionnaires. In order to allow the experts to have more freedom and to express their assessment information precisely, four multi-granularity linguistic term sets dened in Section 3.1 are used in the designed questionnaire, i.e.,
1 1 1 1 1 1 1 1 S 1 fs 1 0 ; s1 ; s2 ; s3 ; s4 ; s5 ; s6 ; s7 ; s8 g; 2 2 2 2 2 2 S 2 fs 2 0 ; s1 ; s2 ; s3 ; s4 ; s5 ; s6 g; 3 3 3 3 S 3 fs 3 0 ; s 1 ; s 2 ; s 3 ; s 4 g; 4 4 4 4 4 4 4 4 S 4 fs 4 0 ; s1 ; s2 ; s3 ; s4 ; s5 ; s6 ; s7 ; s8 g:

Assume the four experts use the four dierent linguistic term sets to express their assessment information over alternatives, where experts D1 ; D2 ; D3 and D4 express their preferences by sets S 1 ; S 2 ; S 3 and S 4 , respectively. The original multi-granularity linguistic information provided by the four experts is presented in the following Table 1. According to model (8), the following linear goal programming model can be set up to obtain the collective ranking values of alternatives. 4 X 4 X 4 1X Minimize z d d kih 4 k1 i1 h1 kih
Subject to pg 1 d 111 d 111 0:37; pv 1 d 114 d 114 0:62; pg 2 d 121 d 121 0:62; pv 2 d 124 d 124 0:87; pg 3 d 131 d 131 0:25; pv 3 d 134 d 134 0:5; u pg pf 4 d 141 d 141 0:5; 4 d 142 d 142 0:62; p 4 d 143 d 143 0:62; pv 4 d 144 d 144 0:75; pg 1 d 211 d 211 0:33; pv 1 d 214 d 214 0:66; pg 2 d 221 d 221 0:5; pf 2 d 222 d 222 0:66; pu 2 d 223 d 223 0:66; pf 1 d 212 d 212 0:5; pu 1 d 213 d 213 0:5; pf 3 d 132 d 132 0:37; pu 3 d 133 d 133 0:37; pf 1 d 112 d 112 0:5; pu 1 d 113 d 113 0:5;

pf 2 d 122 d 122 0:75;

pu 2 d 123 d 123 0:75;

pv 2

d 224

d 224

0:83;

1106

Y.-P. Jiang et al. / Information Sciences 178 (2008) 10981109


pg 3 d 231 d 231 0:33; pv 3 d 234 d 234 0:66; pg 4 d 241 d 241 0:66; pv 4 d 244 d 244 1; pg 1 d 311 d 311 0:25; pv 1 d 314 d 314 0:75; pg 2 d 321 d 321 0:5; pf 2 d 322 d 322 0:75; pu 2 d 323 d 323 0:75; pf 1 d 312 d 312 0:5; pu 1 d 313 d 313 0:5; pf 4 d 242 d 242 0:83; pu 4 d 243 d 243 0:83; pf 3 d 232 d 232 0:5; pu 3 d 233 d 233 0:5;

pv 2 pg 3 pv 3 pg 4 pv 4 pg 1 pv 1 pg 2 pv 2 pg 3 pv 3 pg 4 pv 4

d 324 d 334 d 341 d 344 d 411 d 414 d 421 d 424 d 431 d 434

d 324 d 334 d 341 d 344 d 411 d 414 d 421 d 424 d 431 d 434

1;
pf 3 d 332 d 332 0:5; pu 3 d 333 d 333 0:5;

d 331 d 331 0:25;

0:75; 0; 0: 5; 0:32; 0:65;


0:58; pf 2 d 422 d 422 0:63; pu 2 d 423 d 423 0:8; pf 1 d 412 d 412 0:41; pu 1 d 413 d 413 0:58; pf 4 d 342 d 342 0:25; pu 4 d 343 d 343 0:25;

0:86; 0:17; 0:42; d 443 0:8;


pf 3 d 432 d 432 0:22; pu 3 d 433 d 433 0:36;

d pf pu 4 d 442 d 442 0:63; 441 d 441 0:58; 4 d 443 d 444 d 444 0:86; f f u v u v 0 6 pg 0 6 pg 1 6 p 1 6 p 1 6 p 1 6 1; 2 6 p 2 6 p 2 6 p 2 6 1; f f u v u v 0 6 pg 0 6 pg 3 6 p 3 6 p 3 6 p 3 6 1; 4 6 p 4 6 p 4 6 p 4 6 1; d d and d kih P 0; kih P 0 kih d kih 0; k ; i; h 1; 2; 3; 4:

The above problem is solved by WinQSB 2.0. The obtained collective ranking values of alternatives are ~ p1 0:33; 0:50; 0:50; 0:66; ~ p2 0:58; 0:75; 0:75; 0:87; ~ p3 0:25; 0:50; 0:50; 0:66 and ~ p4 0:58; 0:63; 0:63; 0:86, respectively. By Eq. (9), we have the fuzzy preference relation as follows:

By Eq. (10), we have the non-dominance degree of alternatives, i.e., NDD1 0:242; NDD2 1; NDD3 0:242 and NDD4 0:73. Therefore, the ranking result of the alternatives is X 2  X 4  X 1  X 3 , and thus the best alternatives is X 2 . Next, to illustrate the strengths of the proposed method, we use the fusion method given by Herrera et al. [10] to solve the above GDM problem. The fusion method consists of the following ve steps: (1) Express the multi-granularity linguistic information in the form of fuzzy numbers. (2) Uniform the multi-granularity linguistic information. In this step, the BLTS is chosen as the uniform representation and each linguistic performance value provided by experts is represented as a fuzzy set on the BLTS by means of a transformation function. (3) Obtain the collective ranking values of alternatives by using the OWA aggregation operator. (4) Compute the non-dominance degree of each alternative; and (5) rank the alternatives.

Y.-P. Jiang et al. / Information Sciences 178 (2008) 10981109

1107

By fusion method [10], the collective ranking values of the four alternatives are obtained as follows: r1 0; 0; 0; 0; 0:08; 0:45; 0:72; 0:88; 0:58; 0:31; 0; 0; 0; 0; 0; r2 0; 0; 0; 0; 0; 0; 0; 0; 0:12; 0:82; 0:73; 0:38; 0:1; 0; 0; r3 0; 0; 0; 0:05; 0:23; 0:52; 0:32; 0:17; 0; 0; 0; 0; 0; 0; 0; r4 0; 0; 0; 0; 0; 0; 0; 0; 0; 0:12; 0:27; 0:11; 0; 0; 0: In order to rank all the alternatives and choose the most desirable alternatives, the non-dominance degree of alternatives is used. Firstly, the collective possibility degree of dominance of alternative X i over X j is given as p0ij maxcl minch 6cl flri cl ; lrj ch g, where cl ; ch 2 S T ; lri cl ail and lrj ch aih . Then, the fuzzy preference relation P 0 p0ij nn on the pairwise comparisons of the collective ranking values of alternatives can be constructed [10]. According to the above values r1 ; r2 , r3 and r4 , the fuzzy preference relation is set up as follows: 0 1 0:31 0:52 0:12 B 0:82 0:52 0:27 C B C P 0 p0ij 44 B C: @ 0:45 0 0 A 0:27 0:27 0:27 By Eq. (10), the non-dominance degree of alternatives can be obtained, i.e., NDD01 0:49; NDD02 1, NDD03 0:48 and NDD04 1. Therefore, the ranking result of the alternatives is X 2  X 4  X 1  X 3 . To compare the performances of our method and the fusion method, we introduce the following performance evaluation criterion: m X n X n X k jp 11 DP ij p ij j;
k 1 i1 j 1 k where p ij and p ij denote the collective and individual possibility degree of dominance of alternative X i over X j , respectively. It can be seen that the smaller the value of DP is, the closer the collective opinion to every experts opinion. With respect to the multi-granularity linguistic information provided by experts D1 ; D2 ; D3 and D4 , by Eq. (9), the individual possibility degree of dominance of one alternative over another one is computed and the following fuzzy preference relations can be obtained: 0 1 0 1 0 1 0: 5 0: 5 1 0 B 1 B 1 1 0: 5 C 1 1 C B C B C P 1 p 1 C; P 2 p 2 C; ij 44 B ij 44 B @ 0:5 0 0 A @ 1 0: 5 0 A

0: 5 0: 5 0: 5 0

1 1

B 1 B P 3 p 3 B ij 44 @ 1 0:5

1 1 1 1C C C; 1A

1 0:5833 1 0 1 1 1

0:5833 1 0

1 C C C: A

B 1 B P 4 p 4 B ij 44 @ 0:6667 1

0:5

Considering the collective ranking values of alternatives obtained by the goal programming method and the fusion method, by Eq. (9), we have fuzzy preference relations P pij 44 and P 0 p0ij 44 as before. If p ij in Eq. (11) is substituted by pij and p0ij , we can obtain DP 9:0159 and DP 0 19:71, respectively. A comparison of the results obtained by the two methods is shown in Table 2. It can be noticed from Table 2 that the two collective ranking value vectors obtained by the two methods, T T ~ p1 ; ~ p2 ; ~ p3 ; ~ p4 and r1 ; r2 ; r3 , r4 , are quite dierent, but the two ranking results on the alternatives based on T 1 2 3 4 T ~ p1 ; ~ p2 ; ~ p3 ; ~ p4 and r ; r ; r ; r are similar. Also it can be seen that the performance evaluation criterion values obtained by the two methods, DP and DP 0, are dierent and DP 9:016 < DP 0 19:710: This illustrates that the proposed method has smaller gap than the fusion method between the collective opinion

1108

Y.-P. Jiang et al. / Information Sciences 178 (2008) 10981109

Table 2 Comparative results of the two methods Methods Ranking of alternatives Performance evaluation criterion DP The proposed method The fusion method X2  X4  X1  X3 X2  X4  X1  X3 9.016 19.710

and each experts opinion, and the group consensus of the proposed method is better than that of the fusion method. Besides, the fusion method requires the information transformation, while the proposed method does not require any transformation. As a result, it simplies the computation process and avoids the subjectivity of experts in determining a transformation function. 5. Conclusions This paper presents a new method to solve the GDM problem with multi-granularity linguistic information. The approach is based on a linear goal programming model to integrate the experts assessment information and to directly compute the collective ranking values of alternatives without the need of information transformation. An example is examined to illustrate the potential applications and eectiveness of the proposed method. Its advantages are also analyzed by comparing with the existing approaches [10,12]. The proposed method turns out to be very exible and ecient and provides experts with a new alternative to solve GDM problems with multi-granularity linguistic information. It enriches the theory and methodology of group decision analysis. Acknowledgements This work was partly supported by the National Science Fund for Distinguished Young Scholars of China (No. 70525002), the National Natural Science Foundation of China (Nos. 70721001, 70301008 and 70371050), the Research Fund for the Doctoral Program of Higher Education, Ministry of Education, China (No. 20040145018) and the Key Laboratory of Integrated Automation of Process Industry (Northeastern University), Ministry of Education, China (No. JCLL-01-05). The authors also wish to thank the Editor-in-Chief and the anonymous reviewers for their helpful comments and suggestions on earlier versions of this manuscript. References
[1] G.I. Adamopoulos, G.P. Pappis, A fuzzy linguistic approach to a multicriteria-sequencing problem, European Journal of Operation Research 92 (1996) 628636. [2] P.P. Bonissone, K.S. Decker, Selecting uncertainty calculi and granularity: an experiment in trading-o precision and complexity, in: L.H. Kanal, J.F. Lemmer (Eds.), Uncertainty in Articial Intelligence, North-Holland, Amsterdam, 1986, pp. 217247. [3] G. Bordogna, M. Fedrizzi, G. Pasi, A linguistic modeling of consensus in group decision making based on OWA operators, IEEE Transactions on Systems, Man and Cybernetics Part A: Systems and Humans 27 (1997) 126132. [4] G. Bordogna, G. Pasi, A fuzzy linguistic approach generalizing boolean information retrieval: a model and its evaluation, Journal of the American Society for Information Science 4 (1993) 7082. [5] D.Y. Chang, Applications of the extent analysis method on fuzzy AHP, European Journal of Operational Research 95 (1996) 649 655. [6] S.L. Chang, R.C. Wang, S.Y. Wang, Applying fuzzy linguistic quantier to select supply chain partners at dierent phases of product life cycle, International Journal of Production Economics 100 (2006) 348359. [7] C.T. Chen, Extensions of the TOPSIS for group decision-making under fuzzy environment, Fuzzy Sets and Systems 114 (2000) 19. [8] M. Delgado, J.L. Verdegay, M.A. Vila, On aggregation operations of linguistic labels, International Journal of Intelligent Systems 8 (1993) 351370. [9] D. Dubois, H. Prade, Ranking fuzzy numbers in the setting of possibility theory, Information Sciences 30 (1983) 183224. [10] F. Herrera, E. Herrera-Viedma, L. Martinez, A fusion approach for managing multi-granularity linguistic term sets in decision making, Fuzzy Sets and Systems 114 (2000) 4358.

Y.-P. Jiang et al. / Information Sciences 178 (2008) 10981109

1109

[11] F. Herrera, L. Martinez, A 2-tuple fuzzy linguistic representation model for computing with words, IEEE Transactions on Fuzzy Systems 8 (2000) 746752. [12] F. Herrera, L. Martinez, A model based on linguistic 2-tuples for dealing with multigranularity hierarchical linguistic contexts in multiexpert decision-making, IEEE Transactions on Systems, Man and Cybernetics Part B: Cybernetics 31 (2001) 227234. [13] F. Herrera, L. Martinez, P.J. Sanchez, Managing non-homogeneous information in group decision making, European Journal of Operational Research 166 (2005) 115132. [14] J. Lawry, An alternative approach to computing with words, International Journal of Uncertainty Fuzziness Knowledge-Based Systems 9 (2001) 316. [15] H. Lee, Group decision making using fuzzy sets theory for evaluating the rate of aggregative risk in software development, Fuzzy Sets and Systems 80 (1996) 261271. [16] G.S. Liang, Personnel selection using fuzzy MCDM algorithm, European Journal of Operational Research 78 (1994) 2233. [17] A.I. Olc er, A.Y. Odabas i, A new fuzzy multiple attributive group decision making methodology and its application to propulsion/ manoeuvring system selection problem, European Journal of Operational Research 166 (2005) 93114. [18] J.G. Pachon, C. Romero, Distance-based consensus methods: a goal programming approach, Omega 27 (1999) 341347. [19] G. Pasi, R.R. Yager, Modeling the concept of majority opinion in group decision making, Information Sciences 176 (2006) 390414. [20] M.J. Schniederjans, Goal Programming Methodology and Applications, Kluwer Academic Publishers, 1995. [21] T. Tanino, T. Tanaka, M. Inuiguchi, Multi-Objective Programming and Goal Programming: Theory and Applications, SpringerVerlag, New York, 2003. [22] Y.M. Wang, C. Parkan, Two new approaches for assessing the weights of fuzzy opinions in group decision analysis, Information Sciences 176 (2006) 35383555.

Das könnte Ihnen auch gefallen