Escolar Documentos
Profissional Documentos
Cultura Documentos
c 2005 Springer Science + Business Media, Inc. Manufactured in The Netherlands.
Abstract. Although Yager has presented a prioritized operator for fuzzy subsets, called the non-monotonic oper-
ator, it can not be used to deal with multi-criteria fuzzy decision-making problems when generalized fuzzy numbers
are used to represent the evaluating values of criteria. In this paper, we present a prioritized information fusion
algorithm based on the similarity measure of generalized fuzzy numbers. The proposed prioritized information
fusion algorithm has the following advantages: (1) It can handle prioritized multi-criteria fuzzy decision-making
problems in a more flexible manner due to the fact that it allows the evaluating values of criteria to be represented
by generalized fuzzy numbers or crisp values between zero and one, and (2) it can deal with prioritized information
filtering problems based on generalized fuzzy numbers.
Keywords: prioritized information fusion algorithm, generalized fuzzy numbers, similarity measures, information
filtering, prioritized operator
the symbol “| |” denotes the “absolute” operator. Fur- 2.4. A Ranking Method of Generalized Fuzzy
thermore, because 0 ≤ a1 ≤ a2 ≤ a3 ≤ a4 ≤ 1 Numbers Based on the SCGM [18]
and 0 ≤ b1 ≤ b2 ≤ b3 ≤ b4 ≤ 1, from formulas
(5) and (6), we can see that the values of S Ã and S B̃ From [23, 26, 27], we can see that some centroid-index
are positive values between zero and one, respectively. ranking methods of fuzzy numbers have been presented
Therefore, from formula (4), we can see that the value to calculate the ranking order between fuzzy numbers.
of B(S Ã , S B̃ ) is either zero or one. Because the val- However, according to [18], we can see that there are
ues of x̂ Ã and x̂ B̃ obtained by formula (2), respectively, some drawbacks in the existing methods, i.e., they can-
are between zero and one, and because the values of not correctly calculate the ranking order of the gener-
|ai − bi | and |x̂ Ã − x̂ B̃ | are between zero and one, we alized fuzzy numbers in some situations. Thus, in [18],
can see that formula (3) is not a negative value. In [13], we presented a method to evaluate the ranking order be-
we have shown that the value of S( Ã, B̃) in formula (3) tween generalized fuzzy numbers, based on center-of-
is between zero and one. gravity (COG) points and standard deviations of gen-
The proposed similarity measure of generalized eralized fuzzy numbers. The proposed ranking method
fuzzy numbers has the following properties [13]: can overcome the drawbacks of the existing ranking
methods presented in [23, 26, 27]. Assume that there
Property 2.1. Two generalized fuzzy numbers à and is a generalized fuzzy number Ãi , where Ãi = (ai1 ,
B̃ are identical if and only if S( Ã, B̃) = 1. ai2 , ai3 , ai4 ; w Ãi ), represented as an evaluating value of
an alternative xi , where 1 ≤ i ≤ n. The ranking value
R( Ãi ) of the generalized fuzzy number Ãi is calculated
Property 2.2. S( Ã, B̃) = S( B̃, Ã).
as follows:
In [17], we have pointed out that formula (3) can be where the values of x̂ Ãi and ŷ Ãi are calculated by for-
further simplified. Thus, in [17], we modified formula mulas (1) and (2), respectively, and the value of ŝ Ãi is
(3) into formula (7) to simplify the calculation process the standard deviation of the generalized fuzzy number
shown as follows: Ãi , defined as follows [18]:
4 4
|ai − bi | 1
S( Ã, B̃) = 1− i=1
ŝ Ãi = (ai j − āi )2
4 3 j=1
12
min( ŷ Ã , ŷ B̃ ) 4
×(1 − |x̂ Ã − x̂ B̃ |) × , (7) 1
max( ŷ Ã , ŷ B̃ ) = a 2 − 2āi2 + āi2
3 j=1 i j
where S( Ã, B̃) ∈ [0, 1]. The larger the value of
4
1
S( Ã, B̃), the more the similarity between the gener- = a 2 − āi2
alized fuzzy numbers à and B̃. From the previous dis- 3 j=1 i j
cussions, we can see that the values of |ai − bi | and
4
2
|x̂ Ã − x̂ B̃ | are positive values, respectively, and the val- 1 4
= a −
2
ai j 3 . (9)
ues x̂ Ã and x̂ B̃ obtained by formula (2) are between zero
3 j=1 i j j=1
and one. Therefore, in formula (7), we can see that:
4 The larger the value of R( Ãi ), the better the rank-
|ai − bi |
1− i=1
× (1 − |x̂ Ã − x̂ B̃ |) ≥ 0. ing of Ãi (i.e., xi is the better alternative), where
4 1 ≤ i ≤ n. In [18], we have used the proposed
ranking method for handling the outsourcing targets
In [17], we have shown that the value of S( Ã, B̃) in selection problems. In [28], we have used the pro-
formula (7) is between zero and one. posed ranking method for handling multi-criteria fuzzy
Prioritized Information Fusion Method for Handling Fuzzy Decision-Making Problems 223
T -Norms T -Conorms
decision-making problems based on the FN-IOWA Property 3.1. If Poss[B | A] = 1, then µ D (x) =
operator. µ A (x) ∧ µ B (x). It means that the fuzzy subset A can
be revised by the fuzzy subset B entirely due to the fact
that the fuzzy subset B is completely compatible with
3. A Prioritized Intersection Operator the fuzzy subset A.
Zadeh [29] developed the theory of approximate rea- Property 3.2. If Poss[B | A] = 0, then µ D (x) =
soning. Yager [30] introduced non-monotonic logics µ A (x). It means that the fuzzy subset A can not be
into the theory of approximate reasoning for repre- revised by the fuzzy subset B due to the fact that the
senting default knowledge or commonsense reasoning. fuzzy subset B is completely incompatible with the fuzzy
Furthermore, in [5], Yager presented a prioritized inter- subset A. Thus, we disregard the fuzzy subset B.
section operator, called the non-monotonic intersection
operator η, to emulate common sense reasoning. Let X Yager [33], used the non-monotonic intersection op-
be the universe of discourse, and let A and B be two erator to deal with multi-criteria fuzzy decision-making
fuzzy subsets in X . The non-monotonic intersection problems and presented a type of criterion called a sec-
operator η is defined as follows: ond order criterion. From [3], we can see that the sec-
ond order criterion is a criterion that does not interfere
η(A, B) = D, (10) with our getting satisfactory solution using the first or-
der criterion. Let us consider the following statement
[3]:
where D is also a fuzzy subset of X, such that
“I want a car that is luxurious and inexpensive and
µ D (x) = µ A (x) ∧ (µ B (x) ∨ (1 − Poss[B | A])), if possible I would like to buy it from my brother.”
(11)
The statement reflects a natural language formulation
of a combination of the first order criteria (i.e., a car
where Poss[B | A] is defined as follows: that is luxurious and inexpensive) and the second order
criterion (i.e., if possible I would like to buy it from my
Poss[B | A] = Maxx [µ A (x) ∧ µ B (x)]. (12) brother). Yager [3] pointed out that the criterion A and
the criterion B of formula (11) are called the first order
Dubois and Prade [14] pointed out that Poss[B | A] criterion and the second order criterion, respectively,
denotes the possibility of B given A. Yager [4] summa- and he modified formula (11) into
rized three t-norms (i.e., ∧) and t-conorms (i.e., ∨) as
shown in Table 1, where a ∈ [0, 1] and b ∈ [0, 1]. µ D (x) = µ A (x) ∧ (µ B (x) ∨ (1 − ω(Poss[B | A]))),
Yager [4] pointed out that Poss[B | A] essentially (13)
measures the degree of intersection between the fuzzy
subsets A and B, and 1− Poss[B | A] can be seen as a where ω is a qualifier defined as follows:
measure of conflict between the fuzzy subsets A and B.
filter) for those elements which satisfy the first order numbers are still fuzzy numbers [33]. Assume that there
criterion A. Yager [4] used the operator in fuzzy in- are two trapezoidal fuzzy numbers à = (a1 , a2 , a3 , a4 )
formation fusion structures and he called the operator and B̃ = (b1 , b2 , b3 , b4 ), where 0 ≤ a1 ≤ a2 ≤
“non-monotonic/prioritized intersection operator” be- a3 ≤ a4 ≤ 1 and 0 ≤ b1 ≤ b2 ≤ b3 ≤ b4 ≤ 1.
cause the operator supports a concept of priority, i.e., Let T be a t-norm and let S be a t-conorm, where T:
in η(A, B), A has the priority over B [5]. [0, 1] × [0, 1] → [0, 1] and S : [0, 1] × [0, 1] →
[0, 1]. The t-norm and t-conorm operations between
Example 3.1. Assume that we use the algebraic prod- the trapezoidal fuzzy numbers à and B̃ are shown as
uct (i.e., a ×b) and the algebraic sum (i.e., a +b−a ×b) follows:
to represent t-norms and t-conorms (i.e., ∧ and ∨), re-
spectively. Assume that there are two different criteria
T ( Ã, B̃) = (a1 ∧ b1 , a2 ∧ b2 , a3 ∧ b3 , a4 ∧ b4 ),
A and B, three alternatives x1 , x2 , x3 , and their evalu-
ating values are shown as follows: (15)
S( Ã, B̃) = (a1 ∨ b1 , a2 ∨ b2 , a3 ∨ b3 , a4 ∨ b4 ),
1.0 0.4 0.5 0.2 1.0 0.3
A= , , , B= , , . (16)
x1 x2 x3 x1 x2 x3
We can see that Poss[B | A] = Max [A(x) ∧ B(x)] = where ∧ and ∨ are any t-norm and t-conorm as shown
x
Max [(1.0 ∧ 0.2), (0.4 ∧ 1.0), (0.5 ∧ 0.3)] = 0.4. By in Table 1, respectively. According to formulas (15) and
applying formula (11), we can see that (16), we can see that the results of t-norm and t-conorm
operations of fuzzy numbers are linear. However, the
µ D (x1 ) = 1.0 ∧ (0.2 ∨ (1.0 − 0.4)) = 0.68, traditional t-norm and t-conorm operators with fuzzy
numbers do not preserve the linearity [33]. Thus, in
µ D (x2 ) = 0.4 ∧ (1.0 ∨ (1.0 − 0.4)) = 0.4, [34], Junghanns et al. pointed out that the results of
µ D (x3 ) = 0.5 ∧ (0.3 ∨ (1.0 − 0.4)) = 0.36. t-norm and t-conorm operations of fuzzy numbers us-
ing formulas (15) and (16) are approximate results.
Because µ D (x1 ) > µ D (x2 ) > µ D (x3 ), we can see that In the following, we extend Yager’s prioritized op-
the preferring order of the alternatives is x1 > x2 > x3 . erator to propose a prioritized information fusion algo-
Thus, the best alternative among the alternatives x1 , x2 rithm for aggregating evaluating values represented by
and x3 is x1 . generalized fuzzy numbers. Let X be the universe of
discourse, X = [0, k]. Assume that there are n alter-
natives x1 , x2 , . . . , and xn , and assume that there are
4. A Prioritized Information Fusion Algorithm two different prioritized criteria A and B as shown in
Table 2, where A = ( Ã1 , Ã2 ,. . . , Ãi , . . . , Ãn ) is called
From [1–5], we can see that the prioritized operator the first order criterion, B = ( B̃ 1 , B̃ 2 ,. . . , B̃ i , . . . , B̃ n )
shown in formula (11) is very useful to deal with prior- is called the second order criterion, Ãi and B̃ i are gen-
itized multi-criteria fuzzy decision-making problems. eralized fuzzy numbers, Ãi = (ai1 , ai2 , ai3 , ai4 ; w Ãi ),
However, the prioritized operator shown in formula B̃ i = (bi1 , bi2 , bi3 , bi4 ; w B̃ i ), 0 ≤ ai j ≤ k, 0 ≤ bi j ≤ k,
(11) can not deal with multi-criteria fuzzy decision- 0 < w Ãi ≤ 1, 0 < w B̃ i ≤ 1, 1 ≤ i ≤ n, and 1 ≤ j ≤ 4.
making problems if we use generalized fuzzy numbers The proposed information fusion algorithm is now
to represent evaluating values of criteria due to the fact presented as follows:
that the operator can not measure the degree of inter-
section (i.e., formula (12)) between two sets of gen-
eralized fuzzy numbers. Thus, to develop a prioritized Table 2. Evaluating values of the alternatives using two different
information fusion algorithm for aggregating the evalu- criteria A and B.
ating values represented by generalized fuzzy numbers Alternatives
is one of the important research topics of multi-criteria
fuzzy decision-making problems. Different criteria x1 x2 ··· xi ··· xn
In [31, 32], we can see that t-norms and t-conorm op- First order criterion A Ã1 Ã2 ··· Ãi ··· Ãn
erators can apply to the fuzzy number environment. The Second order criterion B B̃ 1 B̃ 2 ··· B̃ i ··· B̃ n
results of t-norms and t-conorms operations of fuzzy
Prioritized Information Fusion Method for Handling Fuzzy Decision-Making Problems 225
Step 1: If the universe of discourse X = [0, k] and numbers A and B, where A = { Ã∗1 , Ã∗2 , . . . , Ã∗n } and
k = 1, then translate the generalized trapezoidal B = { B̃ ∗1 , B̃ ∗2 , . . . , B̃ ∗n }.
fuzzy numbers Ãi = (ai1 , ai2 , ai3 , ai4 ; w Ãi ) and B̃ i = Step 3: Based on formulas (15), (16) and (19), extend
(bi1 , bi2 , bi3 , bi4 ; w B̃ i ) into standardized generalized formula (11) into
trapezoidal fuzzy numbers Ãi∗ and B̃ i∗ , respectively,
where 1 ≤ i ≤ n, shown as follows:
D̃i = Ãi∗ ∧ ( B̃ i∗ ∨ (1 − C(A, B))), (20)
ai1 ai2 ai3 ai4
Ãi∗ = , , , ; w Ãi∗
k k k k where D̃i is a generalized fuzzy number denoting
∗ ∗ ∗ ∗
= ai1 , ai2 , ai3 , ai4 ; w Ãi∗ , (17) the fusion result of the generalized fuzzy numbers
Ãi∗ and B̃ i∗ , 1 ≤ i ≤ n, and the operators ∧ and ∨ can
bi1 bi2 bi3 bi4
B̃ i∗ = , , , ; w B̃ i∗ be replaced by any t-norm and t-conorms as shown
k k k k in Table 1, respectively. 1-C(A, B) can be seen as a
∗ ∗ ∗ ∗
= bi1 , bi2 , bi3 , bi4 ; w B̃ i∗ . (18) measure of conflict between two sets of generalized
fuzzy numbers A and B. Assume that 1 − C(A, B) =
where 0 ≤ ai∗j ≤ 1, 0 ≤ bi∗j ≤ 1, w Ãi∗ = w Ãi , w B̃ i∗ = p, where p ∈ [0, 1], using formulas (15) and (16) to
w B̃ i , 1 ≤ i ≤ n, and 1 ≤ j ≤ 4. The standardized calculate D̃i as follows:
generalized fuzzy numbers mean that the universes
of discourse U of the generalized fuzzy numbers are D̃i = Ãi∗ ∧ ( B̃ i∗ ∨ (1 − C(A, B))) = Ãi∗ ∧ ( B̃ i∗ ∨ p)
between zero and one. ∗ ∗ ∗ ∗
Step 2: Based on formulas (1), (2) and (7), evaluate the = ai1 , ai2 , ai3 , ai4 ; w Ãi∗
degree of compatibility C(A, B) between the first ∗ ∗ ∗ ∗
order criterion A and the second order criterion B ∧ bi1 , bi2 , bi3 , bi4 ; w B̃ i∗ ∨ p
∗ ∗ ∗ ∗ ∗
shown as follows: = (ai1 ∧ (bi1 ∨ p), ai2 ∧ (bi2 ∨ p), ai3
∧ (bi3∗
∨ p), ai4 ∗ ∗
∧ (bi4 ∨ p);
C(A, B) = Max Min x̂ Ã∗1 ∧ x̂ B̃ ∗1 , S( Ã∗1 , B̃ ∗1 ) ,
w Ãi∗ + w B̃ i∗ × C(A, B)
Min x̂ Ã∗2 ∧ x̂ B̃ ∗2 , S( Ã∗2 , B̃ ∗2 ) ,
1 + C(A, B)
..
. = (di1 , di2 , di3 , di4 ; w D̃i ), (21)
Min x̂ Ãi∗ ∧ x̂ B̃ i∗ , S( Ãi∗ , B̃ i∗ ) , ∗ ∗
w Ã∗ +w B̃ ∗ ×C(A, B)
where dik = (aik ∧(bik ∨ p), w D̃i = i 1+C(A,
i
B)
,
..
. 1 ≤ k ≤ 4, and 1 ≤ i ≤ n. From formula (21), we
∗ can see that the calculation result D̃i is still a general-
Min x̂ Ã∗n ∧ x̂ B̃ ∗n , S( Ãn , B̃ ∗n ) , (19)
ized fuzzy number. In formula (21), we use the mean
w Ã∗ +w B̃ ∗ ×C(A, B)
where value i 1+C(A, i
B)
to represent the fusion result
of the two certainty degrees of the evaluating values
1. the operators ∧ can be replaced by any t-norms w Ãi∗ and w B̃ i∗ in different degrees of compatibility
as shown in Table 1; C(A, B), where C(A, B) ∈ [0, 1].
2. the value x̂ Ãi∗ of the standardized generalized Step 4: Apply formula (8) to obtain the ranking order
fuzzy number Ãi∗ and the value x̂ B̃ i∗ of the stan- R( D̃i ) of the fusion results D̃i , where 1 ≤ i ≤ n.
dardized generalized fuzzy number B̃ i∗ are cal- According to formula (8), the ranking value R( D̃i )
culated by formula (2), respectively; of a generalized fuzzy number D̃i , where D̃i = (di1 ,
3. S( Ãi∗ , B̃ i∗ ) denotes the degree of similarity be- di2 , di3 , di4 ; w D̃i ), can be calculated as follows:
tween the standardized generalized fuzzy num-
bers Ãi∗ and B̃ i∗ calculated by formula (7). ŝ 1−w D̃
R( D̃i ) = x̂ D̃i + w D̃i − ŷ D̃i D̃i × ŷ D̃i + 0.5 i
.
The value of C(A, B) of formula (19) is a real value
between zero and one due to the fact that values x̂ Ãi∗ ∈ (22)
∗
[0, 1], x̂ B̃ i∗ ∈ [0, 1], S( Ãi , B̃ i∗ ) ∈ [0, 1] and 1 ≤ i ≤ n.
The value C(A, B) essentially measures the degree of The larger the value of R( D̃i ), the better the ranking
compatibility between two sets of generalized fuzzy of D̃i , where 1 ≤ i ≤ n. Assume that R( D̃i ) is the
226 Chen and Chen
∗
largest value among the values of R( D̃1 ), R( D̃2 ), . . . Property 4.2. If C(A, B) = 1, then D̃i = Ãi ∧ B̃ i∗ ,
and R( D̃n ), then the alternative xi is the best choice, where 1 ≤ i ≤ n.
where 1 ≤ i ≤ n.
Proof: From formula (21), if C(A, B) = 1, then we
In Step 2 of the proposed algorithm, we use the de- can see that
gree of compatibility C(A, B) (i.e., formula (19)) to
replace Poss[B | A] (i.e., formula (12)). The degree of D̃i = Ãi∗ ∧ ( B̃ i∗ ∨ (1 − C(A, B)))
compatibility C(A, B) has considered two different sit-
∗ ∗ ∗ ∗ ∗
uations shown as follows: = ai1 ∧ (bi1 ∨ 0), ai2 ∧ (bi2 ∨ 0), ai3
Situation 1: If we use the crisp values between zero ∗ ∗ ∗
w Ãi∗ + w B̃ i∗ × 1
∧ (bi3 ∨ 0), ai4 ∧ (bi4 ∨ 0);
and one to represent evaluating values of criteria, 1+1
∗
then the degrees of similarity S( Ã1 , B̃ ∗1 ), S( Ã2 ,
∗
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗
∗ ∗
B̃ 2 ), . . . , S( Ãn , B̃ n ) play a minor role and the re- = ai1 ∧ bi1 , ai2 ∧ bi2 , ai3 ∧ bi3 , ai4
sult of the degree of compatibility C(A, B) is
w Ãi∗ + w B̃ i∗
equal to the result of Poss[B | A] in formula ∗
∧ bi4 ;
(12). 2
Situation 2: If we use the generalized fuzzy numbers to = Ãi∗ ∧ B̃ i∗ ,
represent evaluating values of criteria, then the de-
∗ ∗ ∗
grees of similarity S( Ã1 , B̃ ∗1 ), S( Ã2 , B̃ ∗2 ), . . . , S( Ãn , where the two operators ∧ and ∨ can be replaced by
∗
B̃ n ) play an important role to obtain the degree of any t-norms and t-conorms, respectively, and 1 ≤ i
compatibility C(A, B), due to the fact that the gener- ≤ n.
alized fuzzy numbers have different types of shape,
as discussed in Section 2.
The proposed information fusion algorithm can
handle multi-criteria fuzzy decision-making prob-
In the following, we prove that formula (20) still sat-
lems in a more flexible and more intelligent man-
isfies the two properties of Yager’s non-monotonic in-
ner due to the fact that it allows the evaluating val-
tersection operator (i.e., Property 3.1 and Property 3.2)
ues of the criteria to be represented by generalized
as follows:
fuzzy numbers or crisp values between zero and
∗ one. In the following, we use an example to com-
Property 4.1. If C(A, B) = 0, then D̃i = Ãi , where
pare the proposed information fusion method with
1 ≤ i ≤ n.
Yager’s operator η [5], where the evaluating values
Proof: From formula (21), if C(A, B) = 0, then we are represented by crisp values between zero and
can see that one.
∗
D̃i = Ãi ∧ ( B̃ i∗ ∨ (1 − C(A, B))) Example 4.1. We use the proposed information fu-
sion algorithm to deal with Example 3.1. Assume
∗ ∗ ∗ ∗
= ai1 ∧ (bi1 ∨ 1), ai2 ∧ (bi2 ∨ 1), that the algebraic product (i.e., a × b) and the alge-
braic sum (i.e., a + b − a × b) are used to represent
∗ ∗ ∗ ∗
w Ãi∗ + w B̃ i∗ × 0 the t-norm and the t-conorm (i.e., ∧ and ∨), respec-
∧
ai3 ∨ 1),
(bi3 ∧ai4 ∨ 1);
(bi4
1+0 tively. Assume that there are two different criteria A
∗ ∗ ∗ ∗
w Ãi + 0
∗ and B and assume that there are three alternatives,
= ai1 ∧ 1, ai2 ∧ 1, ai3 ∧ 1, ai4 ∧ 1; x1 , x2 and x3 , where their evaluating values are as
1
∗ ∗ ∗ ∗ follows:
= (ai1 , ai2 , ai3 , ai4 ; w Ãi∗ )
∗
= Ãi , 1.0 0.4 0.5
A= , , ,
x1 x2 x3
where the operators ∧ and ∨ can be replaced by any 0.2 1.0 0.3
B= , , .
t-norms and t-conorms, respectively, and 1 ≤ i ≤ n. x1 x2 x3
Prioritized Information Fusion Method for Handling Fuzzy Decision-Making Problems 227
Table 3. A comparison of the prioritized information fusion method with the prioritized
intersection operator proposed by Yager [5].
Properties
(1.0, 1.0, 1.0, 1.0; 1.0) = (1.0, 1.0, 1.0, 1.0; 1.0)
A= ,
x1 ∧ ((0.2, 0.2, 0.2, 0.2; 1.0) ∨ 0.6)
(0.4, 0.4, 0.4, 0.4; 1.0)
, = (0.68, 0.68, 0.68, 0.68; 1)
x2
= 0.68.
(0.5, 0.5, 0.5, 0.5; 1.0)
x3
In the same way, we can calculate the fusion result D̃2
µ A (x1 ) µ A (x2 ) µ A (x3 ) and D̃3 shown as follows:
= , , ,
x1 x2 x3
Table 4. Evaluating values represented by generalized fuzzy num- in Table 4 into standardized generalized fuzzy numbers
bers of the priority criteria A and B. shown in Table 5.
Alternatives Then, we use the algebraic product (i.e., a × b) and
Different the algebraic sum (i.e., a + b − a × b) as shown in
criteria First order criterion A Second order criterion B
Table 1 to represent the t-norm and the t-conorm (i.e.,
Evaluating values ∧ and ∨), respectively. In the two cases, the alternatives
x1 Ã 1 = (10, 10, 10, 10; 1.0) B̃ 1 = (3, 4, 4, 5; 1.0) x1 , x2 , . . . , x8 can be seen as eight different books and
x2 Ã2 = (0, 0, 0, 0; 1.0) B̃ 2 = (10, 10, 10, 10; 1.0) a decision maker wants to choose a most suitable book
x3 Ã3 = (3, 4, 4, 5; 1.0) B̃ 3 = (3, 3.5, 4.5, 5; 1.0) from these books.
x4 Ã4 = (0, 1, 1, 2; 1.0) B̃ 4 = (1, 2, 2, 3; 1.0)
x5 Ã5 = (10, 10, 10, 10; 1.0) B̃ 5 = (0, 0, 0, 0; 1.0) Case 1: The MCDM framework only considers the first
x6 Ã6 = (3, 4, 4, 5; 1.0) B̃ 6 = (5, 6, 6, 7; 0.9)
order criterion A. Assume that the decision maker’s
statement is as follows:
x7 Ã7 = (4, 5, 5, 6; 1.0) B̃ 7 = (9, 9.5, 10, 10; 1.0)
x8 Ã8 = (0, 1, 1, 2; 0.9) B̃ 8 = (6, 7, 7, 8; 1.0) “I want a book which is talking about fuzzy the-
ory.”
fuzzy numbers to represent the evaluating values of two In this statement, the first order criterion A is “fuzzy
different criteria A and B. Assume that there are eight theory”. According to Table 5, we can see that the
alternatives x1 , x2 , x3 , x4 , x5 , x6 , x7 , x8 and two different books x1 and x5 completely satisfy the first order cri-
criteria A and B, where A is the first order criterion and terion A, the books x3 and x6 have the same satisfying
B is the second order criterion. We use the generalized degrees with respect to the first order criterion A, and
fuzzy numbers to represent the evaluating values Ãi the book x2 does not satisfy the first order criterion
and B̃ i of the alternatives xi , where 1 ≤ i ≤ 8, and the A at all. In this case, because the MCDM framework
generalized fuzzy numbers are defined in the universe only considers the first order criterion A, we use for-
of discourse X = [0, 10] as shown in Table 4. mula (22) to calculate the ranking values R( Ãi∗ ) of
From [3], we can see that there are two multi-criteria other evaluating values Ãi∗ , where the ranking values
decision-making (MCDM) frameworks, i.e., (i) with- R( Ãi∗ ) of the evaluating values Ãi∗ , where 1 ≤ i ≤ 8,
out considering the second order criterion B and (ii) of the first order criterion A are shown in Table 6.
considering the second order criterion B as shown in From Table 6, we can see that the ranking order of
Fig. 3, respectively. these ranking values R( Ãi∗ ) of evaluating values Ãi∗
Yager [3] pointed out that the second order crite- is R( Ã∗1 ) = R( Ã∗5 ) > R( Ã∗7 ) > R( Ã∗6 ) = R( Ã∗3 ) >
rion could be seen as an additional selector or a filter as R( Ã∗4 ) > R( Ã∗8 ) > R( Ã∗2 ). Thus, the preferring or-
illustrated in Fig. 3(ii). In the following, we use two dif- der of these books is as follows:
ferent cases to discuss two different conditions, i.e., (1)
The MCDM framework only considers the first order x1 = x5 > x7 > x6 = x3 > x4 > x8 > x2 ,
criterion A; (2) The MCDM framework both considers
the first order criterion A and the second order crite- and we can see that the most suitable books for the
rion B, and we see the second order criterion B as a decision maker are x1 and x5 .
filter. Firstly, we use formulas (17) and (18) to trans- Case 2: The MCDM framework both considers the first
late the generalized trapezoidal fuzzy numbers shown order criterion A and the second order criterion B,
Figure 3. The MCDM frameworks: (i) without considering the second order criterion B, and (ii) considering the second order criterion B.
Prioritized Information Fusion Method for Handling Fuzzy Decision-Making Problems 229
Alternatives
Different
criteria First order criterion A Second order criterion B
Evaluating values
x1 Ã∗1 = (1.0, 1.0, 1.0, 1.0; 1.0) B̃ ∗1 = (0.3, 0.4, 0.4, 0.5; 1.0)
x2 Ã∗2 = (0.0, 0.0, 0.0, 0.0; 1.0) B̃ ∗2 = (1.0, 1.0, 1.0, 1.0; 1.0)
x3 Ã∗3 = (0.3, 0.4, 0.4, 0.5; 1.0) B̃ ∗3 = (0.3, 0.35, 0.45, 0.5; 1.0)
x4 Ã∗4 = (0.0, 0.1, 0.1, 0.2; 1.0) B̃ ∗4 = (0.1, 0.2, 0.2, 0.3; 1.0)
x5 Ã∗5 = (1.0, 1.0, 1.0, 1.0; 1.0) B̃ ∗5 = (0.0, 0.0, 0.0, 0.0; 1.0)
x6 Ã∗6 = (0.3, 0.4, 0.4, 0.5; 1.0) B̃ ∗6 = (0.5, 0.6, 0.6, 0.7; 0.9)
x7 Ã∗7 = (0.4, 0.5, 0.5, 0.6; 1.0) B̃ ∗7 = (0.9, 0.95, 1.0, 1.0; 1.0)
x8 Ã∗8 = (0.0, 0.1, 0.1, 0.2; 0.9) B̃ ∗8 = (0.6, 0.7, 0.7, 0.8; 1.0)
“I want the most suitable book which is talking Step 3. Based on formulas (15), (16) and (21), we can
about fuzzy theory and if possible I want this calculate the fusion results D̃1 , D̃2 , D̃3 , D̃4 , D̃5 , D̃6 ,
book to be concerning management.” D̃7 and D̃8 , respectively, shown as follows:
In this statement, the first order criterion A is “fuzzy D̃1 = Ã∗1 ∧ ( B̃ ∗1 ∨ (1 − 0.4308))
theory”, the second order criterion B is “management”. = (0.6984, 0.7415, 0.7415, 0.7846; 1.0),
In this case, we apply the proposed information fusion
algorithm to deal with the MCDM framework as fol- D̃2 = Ã∗2 ∧ ( B̃ ∗2 ∨ (1 − 0.4308))
lows: = (0.0, 0.0, 0.0, 0.0; 1.0),
D̃3 = Ã∗3 ∧ ( B̃ ∗3 ∨ (1 − 0.4308))
Step 1. Based on formulas (17), (18) and (22), we can
translate the generalized trapezoidal fuzzy numbers = (0.2095, 0.288, 0.3052, 0.3923; 1.0),
shown in Table 4 into the standardized generalized D̃4 = Ã∗4 ∧ ( B̃ ∗4 ∨ (1 − 0.4308))
fuzzy numbers as shown in Table 5. = (0.0, 0.0655, 0.0655, 0.1397; 1.0),
Step 2. Based on formulas (2), (7) and (19), we can
evaluate the degree of compatibility C(A, B) between D̃5 = Ã∗5 ∧ ( B̃ ∗5 ∨ (1 − 0.4308))
the first order criterion A and the positive criterion = (0.5692, 0.5692, 0.5692, 0.5692; 1.0),
230 Chen and Chen
Table 7. Ranking values R( D̃i ) of each fu- can see that the book x4 is more suitable than the book
sion result D̃i . x8 for the decision maker’s need if we only consider the
Fusion results D̃i Ranking values R( D̃i ) first order criterion A. However, in Case 2, we can see
that the book x8 is more suitable than the book x4 for the
D̃1 1.7273 decision maker’s need if we consider the two different
D̃2 1.0 criteria A and Bsimultaneously. The main reason is that
D̃3 1.2677 the evaluating values Ã∗4 and Ã∗8 of the books x4 and
D̃4 1.0455 x8 , respectively, of the first order criterion A are very
D̃5 1.5692 similar, where Ã∗4 > Ã∗8 ; the evaluating values B̃ ∗4 and
D̃6 1.2933 B̃ ∗8 of the books x4 and x8 , respectively, of the positive
D̃7 1.4561 criterion B are very dissimilar, where B̃ ∗4 < B̃ ∗8 . Thus,
D̃8 1.0523 the preferring order between the books x4 and x8 will
be changed when we consider the two different criteria
A and B simultaneously.
D̃6 = Ã∗6 ∧ ( B̃ ∗6 ∨ (1 − 0.4308))
6. Conclusions
= (0.2354, 0.3311, 0.3311, 0.4354; 0.9699),
D̃7 = Ã∗7 ∧ ( B̃ ∗7 ∨ (1 − 0.4308)) In this paper, we have extended the prioritized operator
= (0.3828, 0.4893, 0.5, 0.6; 1.0), presented in [5] to propose a prioritized information
fusion algorithm based on similarity measures of gen-
D̃8 = Ã∗8 ∧ ( B̃ ∗8 ∨ (1 − 0.4308))
eralized fuzzy numbers for handling prioritized multi-
= (0.0, 0.0871, 0.0871, 0.1828; 0.9699). criteria fuzzy decision-making problems. The prior-
itized multi-criteria fuzzy decision-making problems
where we use the t-norm “a × b” and the t-conorm have two different criteria A and B, where A is the first
“a + b − a × b” shown in Table 1 to denote the order criterion and B is the second order criterion. From
operations of the operators ∧ and ∨, respectively. Section 5, we can see that the second order criterion B
Step 4. By applying formula (22), we can get the rank- can be seen as a filter to choose the most suitable book
ing values R( D̃1 ), R( D̃2 ), R( D̃3 ), R( D̃4 ), R( D̃5 ), from those books which satisfy the first order crite-
R( D̃6 ), R( D̃7 ) and R( D̃8 ) of the fusion results D̃1 , rion A. Furthermore, the proposed information fusion
D̃2 , D̃3 , D̃4 , D̃5 , D̃6 , D̃7 and D̃8, respectively, shown method can handle the decision-making problems in
in Table 7. a more flexible manner due to the fact that it allows
the evaluating values of the criteria to be represented
From Table 7, we can see that the ranking order of by generalized fuzzy numbers or crisp values between
these ranking values is R( D̃1 ) > R( D̃5 ) > R( D̃7 ) > zero and one, where the generalized fuzzy numbers
R( D̃6 ) > R( D̃3 ) > R( D̃8 ) > R( D̃4 ) > R( D̃2 ). Thus, can indicate certainty degrees of the evaluating values
the preferring order of these books is as follows: of the decision maker’s opinions.
The prioritized information fusion method presented
x1 > x5 > x7 > x6 > x3 > x8 > x4 > x2 , in this paper can deal with single-level information
filtering problems. In the future, we will extend the
and we can see that the most suitable book for the proposed prioritized information fusion algorithm to
decision maker is x1 . develop a recursive algorithm for handling multi-level
From the results of Case 1 and Case 2, we can see information filtering problems due to the fact that the
that if we do not consider the second order criterion B decision makers may deal with multi-level information
as a filter, we can not distinguish the preferring order filtering in the real world.
between the books x1 and x5 ; x3 and x6 . However, if
we use the proposed information fusion algorithm to
consider the second order criterion B in the MCDM Acknowledgments
framework, we not only can distinguish the preferring
order between the books x1 and x5 ; x3 and x6 , but also This work was supported in part by the National Sci-
can hold the preferring order of the other books using ence Council, Republic of China, under Grant NSC-
the first order criterion A. Furthermore, in Case 1, we 90-2213-E-011-054.
Prioritized Information Fusion Method for Handling Fuzzy Decision-Making Problems 231
37. D. Dubois and H. Prade, Possibility Theory: An Approach to 40. S. Miyamoto, Fuzzy Sets in Information Retrieval and Cluster
Computerized Processing of Uncertainty, Plenum Press: New Analysis, Kluwer Academic Publishers: Massachusetts, 1990.
York, 1988. 41. D. Ramot, R. Milo, M. Friedman, and A. Kandel, “On
38. M.L. Ginsberg, Nonmonotonic Reasoning, Morgan Kaufman: fuzzy correlations,” IEEE Transactions on Systems, Man, and
Los Altos, California, 1987. Cybernetics- Part B: Cybernetics, vol. 31, no. 3, pp. 381–390,
39. M.M. Gupta, A. Kandel, W. Bandler, and J.B. Kiszka, Approxi- 2001.
mate Reasoning in Expert Systems, North-Holland, Amsterdam, 42. L.A. Zadeh, “Fuzzy sets,” Information and Control, vol. 8,
1985. pp. 338–353, 1965.