Professional Documents
Culture Documents
com
Abstract
In this paper, we study the group decision making problem with linguistic preference relations. We .rst show that the
weighted combination of A1 ; A2 ; : : : ; Am is linguistic preference relation under the condition that all of A1 ; A2 ; : : : ; Am are
linguistic preference relations. Then we de.ne the concepts of deviation degree and similarity degree between two linguistic
values, and deviation degree and similarity degree between two linguistic preference relations. We also show that the deviation
degree between linguistic preference relation Ai of A1 ; A2 ; : : : ; Am and their group linguistic preference relation is no greater
than the largest deviation degree between the linguistic preference relation Ai and each of the linguistic preference relations
A1 ; A2 ; : : : ; Am . Thus, a theoretic basis has been established for the application of linguistic preference relations in group decision
making.
? 2004 Elsevier Ltd. All rights reserved.
Keywords: Linguistic variable; Linguistic preference relation; Deviation degree; Similarity degree; Group decision making; Operational laws
1. Introduction exert the e>ect of the leading decision maker, it is often re-
quired that the group opinion should be close to the one of
Group decision making with linguistic preference rela- the leading decision maker [13]. In this paper, we shall fo-
tions has been investigated by many documents [1–10]. In cus our attention on deviation measures of linguistic prefer-
the process of group decision making under linguistic envi- ence relations given by group members and try to establish
ronment, each decision maker generally needs to compare a a theoretic basis for the application of linguistic preference
set of decision alternatives with respect to a single criterion, relations in group decision making. To do so, the rest of this
and usually constructs a linguistic preference relation by us- paper is organized as follows. Section 2 introduces some
ing some linguistic terms (e.g., when evaluating the speed basic notations and operational laws of linguistic variables.
of a car, linguistic terms like “fast”, “very fast”, “slow” Section 3 gives a representation of group decision making
may be used [4]; when evaluating the “comfort” or “design” problem with linguistic preference relations. Section 4 in-
of a car, terms like “good”, “medium”, “bad” can be used vestigates deviation measures of linguistic preference rela-
[11]). Usually, there arise situations of con;ict and agree- tions. Section 5 gives an illustrative example, and .nally,
ment among preferences of experts. Thus, .nding a group concluding remarks are made in Section 6.
consensus to represent a common opinion of the group is an
important issue. Moreover, there are many applications, es-
pecially situations involving policy speci.cation, which ne- 2. Basic notations and operational laws
cessitate di>erential weights in deciding group preferences
[12], in such cases, there usually exists a leading decision The linguistic approach is an approximate technique,
maker who plays a key role in decision making. In order to which represents qualitative aspects as linguistic values
by means of linguistic variables [14–29]. Suppose that
E-mail address: xu zeshui@263.net (Z. Xu). S = {s | = −t; : : : ; −1; 0; 1; : : : ; t} is a .nite and totally
0305-0483/$ - see front matter ? 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.omega.2004.04.008
250 Z. Xu / Omega 33 (2005) 249 – 254
ordered discrete term set, where s represents a possible Theorem 1. Let A1 ; A2 ; : : : ; Am be the linguistic preference
value for a linguistic variable. For example, a set of nine relations provided by m decision makers dk (k =1; 2; : : : ; m),
terms S could be given as follows: where Ak = (a(k) (k)
ij )n×n ; aij ∈ S (k = 1; 2; : : : ; m; i; j =
1; 2; : : : ; n), then their weighted combination
S = {s−4 = extremely poor; s−3 = very poor;
s0 = fair; s1 = slightly good; is the group linguistic preference relation Â=(âij )n×n , which
satis5es
s2 = good; s3 = very good;
s−t 6 âij 6 st ; âij ⊕ âji = s0 ;
s4 = extremely good}:
âii = s0 ; i; j = 1; 2; : : : ; n:
Usually, in these cases, it is required that there exist the
following: The proof of Theorem 1 is provided in Appendix A.
(1) The set is ordered: s ¿ s if and only if ¿ ;
(2) There is the negation operator: neg(s ) = s− .
4. Deviation measures of linguistic preference relations
To preserve all the given information, we extend the
discrete term set S to a continuous term set SH = {s | ∈ De"nition 1. Let s ; s ∈ S be two linguistic variables, then
[−q; q]}, where q (q ¿ t) is a suIciently large positive in- we de.ne the deviation degree between s and s as follows:
teger. If s ∈ S, then we call s the original term, otherwise,
| − |
we call s the virtual term. d(s ; s ) = ; (2)
2t
Note. In general, the decision maker uses the original lin- where 2t is the number of linguistic terms in the set S.
guistic terms to evaluate alternatives, and the virtual linguis-
tic terms can only appear in operation. Obviously, the smaller the value of d(s ; s ), the smaller
the deviation degree of the linguistic variables s and s .
Consider any two linguistic terms s ; s ∈ S, H and Especially, if d(s ; s ) = 0, then s = s .
;
1 ;
2 ∈ [0; 1], we introduce some operational laws as
follows [10]: Theorem 2. Let s ; s ∈ S, then
s−t 6 a(k)
ij 6 st ; a(k) (k)
ij ⊕ aji = s0 ;
If we let the lower indices of linguistic values correspond-
a(k)
ii = s0 ; i = 1; 2; : : : ; m: ing to aij and bij be I (aij ) and I (bij ) respectively, then (3)
Z. Xu / Omega 33 (2005) 249 – 254 251
The proof of Theorem 5 is provided in Appendix A. S = {s−4 = extremely poor; s−3 = very poor;
d(Â; Ak ) 6 max d(Al ; Ak ) for all k: and construct, respectively, the linguistic preference rela-
l
tions. Suppose that the linguistic preference relation B is
given by a leading decision maker, and the linguistic prefer-
The proof of Theorem 6 is provided in Appendix A.
ence relations A1 ; A2 ; A3 and A4 are given by the other four
decision makers respectively. They are listed as follows:
De"nition 5. Let A and B be two linguistic preference re-
lations, then A and B are of acceptable deviation degree, if s0 s0 s2 s−1 s4
d(A; B) 6 (in general, = 0:15).
s0 s0 s−1 s0 s3
B = s−2 s1 s0 s2 s1 ;
By Theorem 6, we can easily obtain the following
s s0 s−2 s0 s2
corollaries. 1
s−4 s−3 s−1 s−2 s0
Corollary 1. Let A1 ; A2 ; : : : ; Am , and B be m + 1 linguistic
preference relations, and let s0 s−1 s3 s−1 s3
s1 s0 s1 s0 s2
 = 1 A1 ⊕ 2 A2 ⊕ · · · ⊕ m Am
A1 = s−3 s−1 s0 s−1 s2 ;
if d(Al ; B) 6 (l = 1; 2; : : : ; m), then s s0 s1 s0 s0
1
d(Â; B) 6 : s−3 s−2 s−2 s0 s0
252 Z. Xu / Omega 33 (2005) 249 – 254
s0 s1 s2 s0 s4 linguistic preference relation Ai and each of the linguistic
preference relations A1 ; A2 ; : : : ; Am . Especially, a linguistic
s−1 s0 s−1 s0 s0
preference relation B and the collective reference relation of
A2 =
s−2 s1 s0 s−1 s3
;
linguistic preference relations A1 ; A2 ; : : : ; As are of accept-
able deviation degree under the condition that the linguistic
s0 s0 s1 s0 s1
preference relation B and each of the linguistic preference
s−4 s0 s−3 s−1 s0 relations A1 ; A2 ; : : : ; Am are of acceptable deviation degree.
The illustrative example has also veri.ed the developed re-
s0 s0 s3 s1 s3 sults. Thus, a theoretic basis has been established for the ap-
plication of linguistic preference relations in group decision
s0 s0 s−2 s2 s2
making.
A3 =
s−3 s2 s0 s1 s1
;
s−1 s−2 s−1 s0 s−1
Acknowledgements
s−3 s−2 s−1 s1 s0
I am very grateful to the editor and the anonymous
s0 s2 s0 s−1 s2 referees for their valuable comments and suggestions.
This project was supported by China Postdoctoral Science
s−2 s0 s−1 s1 s0
Foundation (2003034366).
A4 =
s0 s1 s0 s−1 s2
;
s1 s0 s1 s0 s1
Appendix A.
s−2 s0 s−2 s−1 s0
Proof of Theorem 1. Since A1 ; A2 ; : : : ; Am are the linguistic
d(A1 ; B) = 0:12 ¡ 0:15; preference relations, it follows that
d(A2 ; B) = 0:11 ¡ 0:15; s−t 6 a(k)
ij 6 st ; a(k) (k)
ij ⊕ aji = s0 ;
6. Concluding remarks = 1 s0 ⊕ 2 s0 ⊕ · · · ⊕ m s0
In this paper, we have shown that the weighted com- = (1 + 2 + · · · + m )s0
bination of A1 ; A2 ; : : : ; Am is linguistic preference relation
= s0 ; i; j = 1; 2; : : : ; n;
under the condition that all of A1 ; A2 ; : : : ; Am are linguistic
preference relations. We have also shown that the devia-
tion degree between the linguistic preference relation Ai of âij ⊕ âji = 1 a(1) (2) (m) (1)
ij ⊕ 2 aij ⊕ · · · ⊕ m aij ⊕ 1 aji
A1 ; A2 ; : : : ; Am and their group linguistic preference relation
is no greater than the largest deviation degrees between the ⊕ 2 a(2) (m)
ji ⊕ · · · ⊕ m aji
Z. Xu / Omega 33 (2005) 249 – 254 253
×(a(m) (m)
ij ⊕ aji )
|I (aij ) − I (cij )|
d(aij ; cij ) =
2t
= 1 s0 ⊕ 2 s0 ⊕ · · · ⊕ m s0
|(I (aij ) − I (bij )) + (I (bij ) − I (cij ))|
=
= (1 + 2 + · · · + m )s0 2t
1 (l) (k)
n n
i.e., = 2 d l aij ; aij
n i=1 j=1
l=1
0 6 d(A; B) 6 1:
m
1 (l)
n n m
Especially, (k)
= 2 d l aij ; l aij
d(A; B) = 0 ⇔ d(aij ; bij ) = 0 for all i; j n i=1 j=1
l=1 l=1
m
⇔ aij = bij for all i; j ⇔ A = B: m (l) (k)
1
n n
l=1 l I (a ij ) − l=1 l I (a )
ij
(2) By (4), we have = 2
n i=1 j=1 2t
1 |I (aij ) − I (bij )|
n n
d(A; B) = 2
n i=1 j=1 2t m (l) (k)
1 l=1 l (I (aij ) − I (aij ))
n n
= 2
n i=1 j=1 2t
1 |I (bij ) − I (aij )|
n n
=
n i=1 j=1
2 2t
1 |I (aij ) − I (aij )|
n n m (l) (k)
6 l
n2 i=1 j=1 2t
= d(B; A): l=1
254 Z. Xu / Omega 33 (2005) 249 – 254