Professional Documents
Culture Documents
Abstract. We extend the ordered weighted averaging (OWA) operator to a provide a new class of
operators called the generalized OWA (GOWA) operators. These operators add to the OWA operator an
additional parameter controlling the power to which the argument values are raised. We look at some
special cases of these operators. One important case corresponds to the generalized mean and another
special case is the ordered weighted geometric operator.
1. Introduction
2. GOWA Operators
X
n
Fða1 ; . . . ; an Þ ¼ wj bj
j¼1
where bj is thePjth largest of the ai and wj are a collection of weights such that
wj 2 [0,1] and nj¼1 wj ¼ 1.
A convenient vector expression of this aggregation operator can be obtained if
we let W be an n-dimension vector whose components are the wj and let B be
an n-dimension vector whose components are the bj . We call W the weighting vec-
tor and B the ordered argument vector. Using these vectors we can express
Fða1 ; . . . ; an Þ ¼ WT B.
By selecting different manifestations of W we can implement different aggrega-
tions. Particularly notable among the operators that can be obtained are the Max,
94 YAGER
Min and the simple average. These are respectively obtained by the vectors W where
w1 ¼ 1 and wj ¼ 0 for j 6¼ 1, W where wn ¼ 1 and wj ¼ 0 for j 6¼ n, and WA where
wj ¼ 1n. Yager (1993) discuses various different examples of weighting vectors.
It has been shown (Yager (1988)) that the OWA operator is a mean operator: it is
symmetric, monotonic and bounded, Mini ½ai Fða1 ; . . . ; an Þ Maxi ½ai . It is also
idempotent, Fða1 ; . . . ; an Þ ¼ a when ai ¼ a for all i.
While the OWA operator can take its arguments values from the real line an
important special case occurs when the arguments are drawn from the unit interval,
I ¼ ½0; 1. In this case F : In ! I. It is this special case we shall focus on.
We now introduce a class of aggregation operator which we shall call the gene-
ralized OWA operators. We shall denote these as GOWA operators.
P
where, wj are a collection of weights satisfying wj 2 ½0; 1 and nj¼1 wj ¼ 1; k is a
parameter such that k 2 ½1; 1; bj is the jth largest of the ai .
Using vector notation we can express this as Mða1 ; . . . ; an Þ ¼ ðWT Bk Þ1=k where W
and B are the vectors introduced earlier. In order to emphasize the parameters W
and k at times we shall indicate this operator as MW=k ða1 ; . . . ; an Þ.
Two special cases are of great significance. First is the case when k ¼ 1, here we get
X
n
Mða1 ; . . . ; an Þ ¼ w j bj ¼ W T B
j¼1
which is the usual OWA operator. The other important special case is when wj ¼ 1n.
In this case
!1=k
X
n
1
Mða1 ; . . . ; an Þ ¼ akj
j¼1
n
This is the generalized mean operator discussed by Dyckhoff and Pedrycz (1984). We
note these are also mean operators: they are symmetric, monotonic and bounded.
Before investigating more special cases we look at some properties of the GOWA
operators. First we see that the GOWA operator is commutative, if P is any per-
mutation then
This implies that the initial indexing of the arguments does not matter.
GOWA AGGREGATION OPERATORS 95
!1=k
X
n
Mða1 ; . . . ; an Þ ¼ wj ak ¼ ðak Þ1=k ¼ a
j¼1
(1) Let Be and B be the associated ordered argument vectors with component b~j and
bj respectively. Since a~i aiit is the case ~
Pn 1=kthat bj bj for all j.
k
(2) Consider now the term f ¼ j¼1 wj bj we now Pshow thatit is monotonic in
1 n k
bj . First we take the natural log of f, log ½f ¼ k log j¼1 wj bj and then we take
the derivative with respect to bj
of
since d db
log f
j
0 then ob j
0. Furthermore since bj is monotonic with respect to the ai
then the result follows.
The boundness of Mða1 ; . . . ; an Þ can easily be established. Since aj Maxi ½ai ¼ a
from
P the monotonicity
1=k we get Mða1 ; . . . ; an Þ Mða ; . . . ; a Þ. Since Mða ; . . . ; a Þ ¼
n k
j¼1 wj a ¼ a then Mða1 ; . . . ; an Þ Maxj ðaj Þ. Similarly we can show that
Mða1 ; . . . ; an Þ Mini ½ai . Thus we see that the GOWA operator is bounded.
The satisfaction of these properties, commutativity, boundedness and monoto-
nicity implies that the GOWA operators are mean operators for any choice of k and
W.
An additional property associated with the GOWA operators is monotonicity
with respect to k, if k > k0 then
!1=k !1=k0
X
n X
n
0
wj bkj wj bkj
j¼1 j¼1
96 YAGER
The proof of this is essentially the same as the proof that the generalized mean has
this property (Dyckhoff and Pedrycz (1984)).
It is also the case that this operator exhibits a monotonicity with respect to the
vector W. In particular, as more of the weight is allocated to components higher in
W the value of the aggregation increases. Formally we express this as follows. Let k1
and k2 be two indices such that k1 > k2 . If W and W e are two weighing vector such
that
then
Thus in summary the GOWA operator is a mean operator that is monotonic with
respect to both its parameters, W and k.
We now look at some special cases obtained by using different choices of the
parameters W and k.
First we consider some cases of W. If W ¼ W where w1 ¼ 1 and wj ¼ 0 for all
j 6¼ 1 then
!1=k
X
n
Mða1 ; . . . ; an Þ ¼ wj bkj ¼ ðbk1 Þ1=k ¼ b1 ¼ Maxi ½ai
j¼1
Thus here with W ¼ W we always get the Max independent of the selection of k. In
the case where W ¼ W where wn ¼ 1 and wj ¼ 0 for all j 6¼ n we can show that
Mða1 ; . . . ; an Þ ¼ Mini ½aj independent of the selection of k. More generally if W½k is a
focused weighting vector having wk ¼ 1 and wj ¼ 0 for all j 6¼ k then for any k we get
Mða1 ; . . . ; an Þ ¼ bk , the kth largest of the arguments. In this case the aggregation is
effectively based on only one argument. More generally note that if wk ¼ 0 then
0 11=k
B C
B P C
B n C
Mða1 ; . . . ; an Þ ¼ B
B
wj bkj C
C
Thus if the kth weight in W is zero then the kth largest
@j ¼ 1 A
j 6¼ k
argument is disregarded in the aggregation. Here the kth largest argument plays in
the ordering process but not in the actual calculation.
We already noted in the special case where wj ¼ 1n for all j we get the generalized
mean.
Another important special case occurs when w1 ¼ a and wn ¼ 1 a. This corre-
sponds to the Hurwicz weighting vector WH . In this case
GOWA AGGREGATION OPERATORS 97
where b1 ¼ Maxi ½ai and bn ¼ Mini ½ai . An interesting special case is when a ¼ 1=2.
Here we get
1=k
1
Mða1 ; . . . ; an Þ ¼ ðbk1 þ bkn Þ1=k
2
Let us now consider the form of the GOWA operator for some particular cases of
k. As we have already noted when k ¼ 1 we get the usual OWA operator. When
k ¼ 2 we get
!1=2
X
n
Mða1 ; . . . ; an Þ ¼ wj b2j
j¼1
Qn
w1 wn 1 1 j¼1 bj
Mða1 ; . . . ; an Þ ¼ þ þ P
¼ n wj ¼
b1 bn j¼1 bj
P n Qn
j¼1 wj bi
i¼1
i 6¼ 1
Q
n
This is closely related to the Harmonic average. If we denote Prodj ¼ bi then
i¼1
Qn i 6¼ j
j¼1 bj
Mða1 ; . . . an Þ ¼ Pn
j¼1 wj Prodj
Since the bj are indexed in decreasing order we see that for j < i we have
Prodj Prodi . From this we see that as the weights move to the lower elements the
value of the resulting aggregation increases since its denominator increases while the
numerator remains the same.
Consider now the case where k ! 0. In this case we get
Y
n
w
Mða1 ; . . . ; an Þ ¼ bj j
j¼1
We observe that this is closely related to the geometric mean. This special case with
k ! 0 has been studied in Chiclana et al. (2000) and Xu and Da (2002) where it was
98 YAGER
called this the ordered weighted geometric (OWG) operator. In Herrera et al. (in
press) the authors have indicated its usefulness in decision making in the case where
the criteria are measured on ratio scales.
We note the following property for GOWA operators with k ! 0:
wn
Qn wj We see this since if 9 aj ¼ 0 then bn ¼ 0 and since wn 6¼ 0 then bn ¼ 0 thus
Proof:
j¼1 bj ¼ 0.
This leads us to observe an important property of these GOWA operators.
THEOREM For any GOWA operator with k < 0 and having a W for which wn 6¼ 0 any
aggregation MW=k ða; . . . ; an Þ in which one argument has value, zero results in
Mða1 ; . . . ; an Þ ¼ 0.
Proof: We have just show that this holds for k > 0. The monotonicity of the
GOWA with respect to k implies the property holds for all k < 0.
We now consider the case in which k ! 1. Here then we have
!1=k
X
n
MW=k ða1 ; . . . ; an Þ ¼ Lim wj bkj ¼ Max ½bj :
k!1 all j s:t: wj 6¼0
j¼1
Thus here we get as the aggregated value the largest argument which has a non-zero
weight. Since the bj are in descending then MW=k ða1 ; . . . ; an Þ ¼ b1 if w1 6¼ 0. It is
interesting to note that if k ! 1 but W is such that wn ¼ 1, wj ¼ 0 for all j 6¼ n, then
MW=k ða1 ; . . . ; an Þ ¼ Mini ½ai ¼ bn :
In the case where k ! 1. we get MW=k ða1 ; . . . ; an Þ ¼ Minall j s:t wj 6¼0 ½bj . Here we
get the smallest argument which has a non-zero weight. In particular if wn 6¼ 0 then
Mða1 ; . . . ; an Þ ¼ bn .
However if W is such that w1 ¼ 1 then even though k ! 1 we get
Mða1 ; . . . ; an Þ ¼ Maxi ½ai ¼ b1 .
We earlier noted the special case in which our weights are of the Hurwicz type WH
(Hurwicz (1951)). In this case w1 ¼ a. and wn ¼ ð1 aÞ and the aggregated value is
b1 bn
Mða1 ; . . . ; an Þ ¼
abn þ ð1 aÞb1
GOWA AGGREGATION OPERATORS 99
oM b2n a oM b21 a
¼ and ¼
ob1 ðabn þ ab1 Þ 2 obn ðabn þ ab1 Þ2
We note that since b1 > bn there is a tendency for the smaller value, bn , to have a
larger derivative and hence be more influential in the aggregation, this effect is
of course modulated by the value of a. As a matter of fact oM oM
ob1 ¼ obn when
2
b2n a ¼ b21 ð1 aÞb2n a ¼ b2n a a
hence when 1a ¼ bb1n .
We note we additional assume a ¼ 1=2 then Mða1 ; . . . ; an Þ ¼ 1ðbb1þb
bn
Þ
¼ b2b1 þb
1 bn
2
.
2 1 2
Yager (1988) associated with the OWA operator a measure called the attitudinal
character of the aggregation. This measure of attitudinal character is a number in the
unit interval indicating the ‘‘Andness/Orness’’ or equivalently the Miness/Maxness
of the aggregation. Essentially it provides a scalar valued characterization of the type
of aggregation being performed. For the ordinary OWA operator the attitudinal
character, which is just dependent on W, was defined as
X
n
nj
A CðWÞ ¼ wj
j¼1
n1
note that W and W are dual. We also note that a symmetric weighting vector is self
dual, We ¼ W. Let us look at the relationship between A–CðWÞ and A–CðWÞ. ~
1 X n
ACðWÞ ¼ wj ðn jÞ
n 1 j¼1
e ¼ 1 X n
1 X n
ACðWÞ w~j ðn jÞ ¼ wnjþ1 ðn jÞ
n 1 j¼1 n 1 j¼1
e ¼ 1 X i
1 X n
ACðWÞ wi ðn ðn i þ 1ÞÞ ¼ wi ði 1Þ
n 1 i¼n n 1 i¼1
e ¼ 1 X n
1 X n
ACðWÞ wi ði nÞ þ ðn 1Þ
n 1 i¼1 n 1 i¼1
e ¼ 1 ACðWÞ
ACðWÞ
Thus the attitudinal character of dual weighting vectors are complements of each of
other.
Yager (1988) suggested an interpretation of the attitudinal character of the
aggregation that allows us to extend it to GOWA operators having parameters W
and k. In Yager (1988) it was noted this attitudinal character is the OWA aggre-
1 0
gation of the argument n1 ; n2
n1 n1 ; . . . ; ;
n1 n1 , A–CðWÞ ¼ WT B where B has com-
nj
ponents bj ¼ n1. Using this we can define the attitudinal character of the GOWA
operator
X
n k !1=k
nj
ACðW=kÞ ¼ wj ¼ ðWT Bk Þ1=k
j¼1
n1
P 1=k
1 n k
We can also express this as A–CðW=kÞ ¼ n1 j¼1 wj ðn jÞ . It is the GOWA
n1 n2 1 0
aggregation of the linear argument n1 ; n1 ; . . . ; n1 ; n1 :
Let us now obtain the attitudinal character for some examples of the GOWA
Qn nj First consider the class where k ! 0 here we nn
operators. get A–CðW=kÞ ¼
j¼1 n1 wj . If we additional assume that w n ¼
6 0, since b n ¼ n1 ¼ 0, we get A–
CðW=kÞ ¼ 0. As we have previously indicated the GOWA is monotonic with respect
GOWA AGGREGATION OPERATORS 101
Since wn ¼ 1n from the preceding observation we have that A–CðWA =kÞ ¼ 0 for k 0.
We now investigate what happens to A–CðWA =kÞ when k >0 (see Figures1 and 2).
1
11=k Pn k
1=k
In Figure 1, we have plotted A–CðWA =kÞ ¼ n1 n j¼1 ðn jÞ as a
function of n for k ¼ 20, 15, 10, 4, 2. We observe that A–CðWA =kÞ leads to be higher
for smaller n, although not significantly, and it asymptotically approaches some limit
which depends on k. The bigger k the closer the limit is to one.
In Figure 2, we have
Pplotted these 1=kasymptotic limits for n ¼ 100. Here we plotted
1 1 100 k
A–CðWA =kÞ ¼ 99 100 j¼1 ðn jÞ for k=1–30. We observe that A–CðWA =kÞ
increases as k increases going from A–CðWA =kÞ ¼ 0:5 for k ¼ 1 to A–C ðWA =kÞ ¼
0:9 for k ¼ 30.
We now consider the case where W ¼ W½k , here wk ¼ 1. Let us see the effect of k.
Here
!1=k
1 Xn
k 1 nk
ACðW½k =kÞ ¼ wj ðn jÞ ¼ ððn kÞk Þk ¼
n 1 j¼1 n1 n1
Figure 1.
102 YAGER
Figure 2.
What is interesting is that A–CðW½k =kÞ is the same for all k, it just depends on k.
We now turn to symmetric weighting vectors. We previously noted that for a
symmetric weighting vector W we get A–CðWÞ ¼ 0:5. This is not necessarily the case
when k 6¼ 1. In the following we consider the special case of symmetric vector where
wj ¼ wjnþ1 ¼ 0:5 for some j. For the case where n ¼ 100 and j ¼ 2 in Figure 3 we
plot A–CðW=kÞ for k ¼ 20 to 20. We see that as k gets smaller we have A–
CðW=kÞ ! 0, we get a kind of Min aggregation. On the other hand when k gets
larger we have A–CðW=kÞ ! 1 giving us a more Max like aggregation.
In Figure 4, we consider the situation of different values of j in the above sym-
metric weighting vector. We plot A–CðW=kÞ for k ¼ 30 to 30 and for j ¼ 2, 20, 30
and 49 (in increasing thickness). We see that as the two symmetric weights move
closer to the center, j getting larger, this operator acts more like A–CðW=kÞ ¼ 0:5.
Figure 3.
GOWA AGGREGATION OPERATORS 103
Figure 4.
where b1 ¼ 1 and bn ¼ 0. We see that for k < 0 this has value zero. For k > 0 we
have
Consider now the situation for 0 < k < 1. We see that for k ¼ 1 we get
ACðWH =kÞ ¼ a. Increasing k, letting it go to 1, leads us to obtaining
ACðWH =kÞ ! 1. On the other hand decreasing k, letting go to zero results in
having ACðWH =kÞ ! 0. This situation very clearly displays the effect of k.
Yager (1996) discusses various different methods for obtaining the OWA weighing
vectors. One important method for generating the weights is via a function
f : ½0; 1 ! ½0; 1 for which fð0Þ ¼ 0, fð1Þ ¼ 1 and fðxÞ ‡ fðyÞ if x > y. These functions
are called basic unit interval monotonic (BUM)
functions. Using these functions we
can generate the OWA weights as wi ¼ f ni fði1 n Þ. Yager (1996) also discusses how
these BUM functions can be related to linguistic concepts. This facility plays a
important role in implementing Zadeh’s paradigm of computing with words (Zadeh
(1996), Zadeh (1999), Zadeh and Kacprzyk (1999)) by providing us with an ability to
induce information aggregation procedures guided by verbally expressed concepts.
104 YAGER
Using these BUM functions in the GOWA aggregation we get the formulation
!1=k !1=k
X
n Xn
j j1
Mða1 ; . . . ; an Þ ¼ wj bkj ¼ f f bkj :
j¼1 j¼1
n n
An important special case of the above occurs when fðxÞ ¼ xP where, P 2 ½0; 1.
We shall denote these as Wp =k. It is note that if addition we assume k ¼ 1 we have
X P !
n
j j1 P
Mða1 ; . . . ; an Þ ¼ bj :
j¼1
n n
P Operator AC
P!0 Max 1
P!1 Average 0:5
P!1 Max 0
We note for this case where P ¼ 1 as k goes from 1 to 1; M goes from Min to
Max. Thus here for P ¼ 1 ACðWp =kÞ goes from 0 to 1 as k goes from 1 to 1:
Let us look at ACðWp =kÞ for other values of k and P. In Figure 5, we plot
ACðWp =kÞ for P ¼ 0:02, 0.2, 1, 2, 7, 20 and for k ¼ 30 to 30 with n ¼ 100.
The use of a weight generating function allows us to very naturally introduce
importance weights into the GOWA aggregation process. Assume associated with
each argument we have an importance weight uj . We shall assume that these weights
lie in the unit interval although we need not assume that they sum to one. Without
lose of generality we shall assume that the aj have been indexed in descending order
aj ¼ bj . Here then uj is the importance weight associated with bj . Using these im-
portances and theP function f we getP
the weight wj associated with bj in the following
way. We let Sj ¼ ji¼1 ui and T ¼ ni¼1 ui then we obtain
Sj Sj1
wj ¼ f f
T T
GOWA AGGREGATION OPERATORS 105
There exists the possibility of applying the ideas introduced here to situations in
which we use more sophisticated methods to generate the weighting vector. One such
situation is where we use a fuzzy measure to generate the weighting vector (Sugeno
(1977)). Implementing this leads to an extension of the Choquet integral (Sugeno and
Murofushi (1987)) to what we shall call the Generalized Choquet Aggregation. Let us
look at this extension.
Here we assume a set of objects A ¼ fA1 ; . . . ; An g. Associated with each object Ai
is a value which desire to aggregate, we shall denote these values as ai and refer to
them as the argument values. In addition we have a measure l on subsets of A,
l : 2A ! ½0; 1, such that lðHÞ ¼ 0, lðAÞ ¼ 1 and lðFÞ ‡ lðEÞ for E F. It is
interesting to note the similarity between the measure l and the BUM function f
introduced previously. In this framework we use the measure l to generate the
weights. We shall find it convenient to introduce function index such that index(j) is
the index of the jth largest of ai . Thus aindexðjÞ is the jth largest argument. We now
define Hj ¼ fAindexðkÞ jk ¼ 1 to jg, it is the subset of j objects having the largest
argument values. Using this we define the set of weights
106 YAGER
wj ¼ lðHj Þ lðHjþ1 Þ
It is easy to show that the wj 2 ½0; 1 and that they sum to one.
The normal Choquet integral of the argument collection ða1 ; a2 ; . . . ; an Þ with re-
spect to the measure l is defined as
X
n
Cl ða1 ; a2 ; . . . ; an Þ ¼ wj aindexðjÞ
j¼1
Using this structure and the ideas introduced in this work we can provide an
extension to a Generalized Choquet integral. Letting k 2 ½1; 1 we define the
Generalized Choquet integral as
!1=k
X
n
Cl=k ða1 ; a2 ; . . . ; an Þ ¼ wj akindexðjÞ
j¼1
7. Conclusion
We extended the OWA operator to a provide a new class of operators called the
GOWA operators with addition of a parameter controlling the power to which the
argument values are raised. We looked at some special cases of these operators in
order to get some understanding of this operator. We studied the attitudinal char-
acter of these operators. A future step that must be taken is the extension of the
method for learning the OWA weights of (Filev and Yager (1998)) to this more
general case where we must also learn the parameter k.
References
Chiclana, F., F. Herrera, and E. Herrera-Viedma. (2000). ‘‘The Ordered Weighted Geometric Operator:
Properties and Applications,’’ Proc. of 8th Int. Conference on Information Processing and Management
of Uncertainty in Knowledge-based systems, Madrid, 985–991.
Dyckhoff, H. and W. Pedrycz. (1984). ‘‘Generalized Means as Model of Compensative Connectives,’’
Fuzzy Sets and Systems 14, 143–154.
Filev, D. P. and R. R. Yager. (1998). ‘‘On the Issue of Obtaining OWA Operator Weights,’’ Fuzzy Sets and
Systems 94, 157–169.
Herrera, F., E. Herrera-Viedma, and F. Chiclana. (In press). ‘‘A Study of the Origins and Uses of the
Ordered Weighted Geometric Operator in Multicriteria Decision Making,’’ International Journal of
Intelligent Systems.
Hurwicz, L. (1951). ‘‘Optimality Criteria for Decision Making Under Ignorance,’’ Cowles Communication
Discussion Paper, Statisticsm No. 370.
Sugeno, M. (1977). ‘‘Fuzzy Measures and Fuzzy integrals: A survey’’. In M. M. Gupta, G. N. Saridis, and
B. R. Gaines (eds.), Fuzzy Automata and Decision Process. Amsterdam: North-Holland Pub, 89–102.
GOWA AGGREGATION OPERATORS 107
Sugeno, M. and T. Murofushi. (1987). ‘‘Choquet Iintegral as an Integral Form for a Class of Fuzzy
Measures,’’ Proceedings of the Second IFSA Congress, Tokyo, 408–411.
Xu, Z. S. and Q. L. Da (2002). ‘‘The Ordered Weighted Geometric Averaging Operator,’’ International
Journal of Intelligent Systems 17, 709–716.
Yager, R. R. (1988). ‘‘On Ordered Weighted Averaging Aggregation Operators in Multi-criteria Decision
Making,’’ IEEE Transactions on Systems, Man and Cybernetics 18, 183–190.
Yager, R. R. (1993). ‘‘Families of OWA operators,’’ Fuzzy Sets and Systems 59, 125–148.
Yager, R. R. (1996). ‘‘Quantifier Guided Aggregation Using OWA Operators,’’ International Journal of
Intelligent Systems 11, 49–73.
Yager, R. R. and J. Kacprzyk. (1997). The Ordered Weighted Averaging Operators: Theory and Applica-
tions. Norwell, MA: Kluwer.
Zadeh, L. A. (1996). ‘‘Fuzzy Logic = Computing with Words,’’ IEEE Transactions on Fuzzy Systems 4,
103–111.
Zadeh, L. A. (1999). ‘‘From Computing with Numbers to Computing with Words-From Manipulation of
Measurements to Manipulations of Perceptions,’’ IEEE Transactions on Circuits and Systems 45, 105–
119.
Zadeh, L. A. and J. Kacprzyk. (1999). Computing with Words in Information/Intelligent Systems 1, Hei-
delberg: Physica-Verlag.