INFORMATICAInformatica1822-88440868-49520868-4952Vilnius UniversityINFO116710.15388/Informatica.2018.162Research ArticleAn Approach to Decision Making with Interval-Valued Intuitionistic Hesitant Fuzzy Information Based on the 2-Additive Shapley FunctionZhangLifeiifei@hnu.edu.cn1
L. Zhang received her PhD degree in corporate management from Hunan University in 2007. Currently, she is an associate professor in Hunan University. She has contributed over 40 journal articles to professional journals such as China Soft Science, China Industrial Economics, Studies in Science of Science, Journal of Management Sciences in China, R&D Management, Systems Engineering-Theory & Practice, Statistics and Decision, Forum on Science and Technology in China, Soft Science, Journal of Intelligence. Her current research interests include alliance governance, decision making and game theory.
TangJietjie411@126.com2
J. Tang is carrying out her PhD degree in School of Business, Central South University. She received her master degree in School of Management, Qingdao Technological University, China, in 2008. Currently, she has contributed over 10 journal articles to professional journals including International Journal of Intelligent Systems, Computers & Industrial Engineering, International Journal of Information Technology & Decision Making, Soft Computing, International Journal of Fuzzy Systems, International Transactions in Operational Research, Neural Computing and Applications, International Journal of Computational Intelligence Systems, Systems Engineering-Theory & Practice. Her research interests include uncertain multi-attribute decision making, cluster analysis and pattern recognition.
MengFanyongmengfanyongtjie@163.com23∗
F. Meng received his PhD degree in management science and engineering from Beijing Institute of Technology in 2011. Currently, he is a professor in in Central South University. He has contributed over 100 journal articles to professional journals such as Omega, Applied Mathematics and Computation, Group Decision and Negotiation, Information Fusion, Information Sciences, Applied Mathematical Modelling, Applied Soft Computing, Knowledge-Based Systems, Computers & Industrial Engineering, IEEE Transactions on Systems, Man, and Cybernetics Systems, Applied Mathematics Letters, Fuzzy Optimization and Decision Making, Soft Computing, International Journal of Fuzzy Systems, Pattern Analysis and Applications, Cognitive Computation, Informatica, Technological and Economic Development of Economy, Operations Research Letters, Journal of the Operational Research Society, Operational Research, etc. His current research interests include fuzzy mathematics, decision making, and game theory.
Interval-valued intuitionistic hesitant fuzzy sets (IVIHFSs) are useful to denote the decision makers’ interval preferred, interval non-preferred and hesitant opinions simultaneously. Considering the application of IVIHFSs, this paper introduces a new decision-making method with interval-valued intuitionistic hesitant fuzzy information that extends the application scopes. To do this, the interval-valued intuitionistic hesitant fuzzy hybrid Shapley weighted averaging (IVIHFHSWA) operator and the interval-valued intuitionistic hesitant fuzzy hybrid Shapley weighted geometric (IVIHFHSWG) operator are defined to aggregate the collective attribute values of alternatives. To reflect the interactions and reduce the complexity of calculating the weights, the 2-additive measures are used to define these two hybrid Shapley weighted operators. To derive the exact weight information of attributes and ordered positions, the associated programming models for determining the optimal 2-additive measures are constructed that are based on the defined Hamming distance measure. To show the feasibility and efficiency of the new method, a practical decision-making problem is offered, which is also used to compare with the previous methods.
With the socioeconomic development, the complexity of decision-making problems is constantly increasing. To denote the fuzzy and uncertain information in decision making, researchers applied fuzzy sets introduced by Zadeh (1965) to cope with this situation. Later, Atanassov (1983) noted that fuzzy sets can only express the decision makers’ preference information and introduced the concept of intuitionistic fuzzy sets (IFSs), which are expressed by two real values in [0,1] to denote the membership and non-membership degrees, respectively. However, due to various kinds of reasons, it is not an easy thing to give the exact values of the membership and non-membership degrees. Thus, Atanassov and Gargov (1989) further gave the concept of interval-valued intuitionistic fuzzy sets (IVIFSs) that apply two intervals in [0,1] to give the uncertain membership and non-membership information. Following the original works of Atanassov (1983) and Atanassov and Gargov (1989), the theory and application of decision making with (interval-valued) intuitionistic fuzzy sets are developed and becomes a hot researching topic (Beliakov and Janes, 2013; Chen and Huang, 2017; Gou and Xu, 2017; Garg, 2017; Liu et al., 2017a; Meng and Chen, 2017; Meng et al., 2017d; Ureña et al., 2015; Wang and Chen, 2017).
Recently, Torra (2010) found that there might be several values for a judgment rather than only one. To address this issue, Torra (2010) introduced the definition of hesitant fuzzy sets (HFSs), which are denoted by several values in [0,1]. To endow the decision makers with more rights to denote their hesitant information, Chen et al. (2013) proposed interval-valued hesitant fuzzy sets (IVHFSs) that use several intervals in [0,1] to express the decision makers’ uncertain hesitancy. To calculate the collective attribute values, the authors introduced two aggregation operators: the generalized interval-valued hesitant fuzzy hybrid averaging (IVHFWA) operator and the generalized interval-valued hesitant fuzzy hybrid geometric (GIVHFWG) operator. Furthermore, He et al. (2016) developed an approach to group decision making with interval-valued hesitant fuzzy information using the interval-valued hesitant fuzzy weighted power Bonferroni mean (IVHFWPBM) operator. Jin et al. (2016) studied the cross-entropy and similarity measures of interval-valued hesitant fuzzy elements (IVHFEs) and illustrated their application for evaluating emergency risk management (ERM). Note that the cross-entropy and similarity measures in Jin et al. (2016) require the considered IVHFEs to have the same length; otherwise, it needs to extend the IVHFEs with the shorter length. Meng et al. (2016) considered the correlation coefficients of IVHFEs, which permit IVHFEs to have the different numbers of intervals. To address the situations where the weights of attributes are interactive, Meng and Chen (2014) defined the induced generalized interval-valued hesitant fuzzy hybrid Shapley averaging (IG-IVHFHSWA) operator and the induced generalized interval-valued hesitant fuzzy hybrid Shapley geometric mean (IG-IVHFHSGM) operator, which are general cases of the Chen et al.’s aggregation operators.
From the concepts of interval-valued intuitionistic fuzzy sets and interval-valued hesitant fuzzy sets, one can find that the former denotes the interval preferred and non-preferred degrees of the decision makers, while the latter indicates the decision makers’ hesitancy. However, they cannot express the decision makers’ interval preferred, interval non-preferred and hesitant opinions simultaneously. Thus, Zhang (2013) presented interval-valued intuitionistic hesitant fuzzy elements (IVIHFEs) that use several interval-valued intuitionistic fuzzy values (IVIFVs) (Xu and Chen, 2007) to denote the decision makers’ opinions. Then, the author defined a series of aggregation operators to rank objects. Later, Joshi and Kumar (2016) developed a method to multi-criteria decision making with interactive characteristics, which is based on the interval-valued intuitionistic hesitant fuzzy Choquet integral (IVIHFCI) operator with respect to the λ-fuzzy measures and the hamming distance on IVIHFEs. The IVIHFCI operator can be seen as an extension of the ordered weighted averaging (OWA) operator in the setting of interactions. However, the Yager’s OWA operator is associated with the weights of the ordered positions, while the IVIHFCI operator considers the weights of the criteria. This means that the objects’ comprehensive criteria values calculated by using the IVIHFCI operator are based on the different criteria weight vectors, namely, the different evaluation standards are used for the different objects. Furthermore, the λ-fuzzy measures can only reflect the complementary or redundant interactions between elements in a set, but they cannot deal with these two aspects simultaneously. Nevertheless, when the importance of the elements is interdependent, we cannot guarantee that there are only complementary or redundant interactions between them. Moreover, the hamming distance on IVIHFEs is unreasonable either. Just as Meng et al. (2016) noted, it is not suitable to extend the shorter IVIHFE to the length of the longer IVIHFE by adding some IVIFV several times since it drives a different IVIHFE. Note that the methods in Zhang (2013), Joshi and Kumar (2016) are both based on the operational laws in (Xu and Da, 2002), which have some undesirable properties.
This paper continues to study decision making with interval-valued intuitionistic hesitant fuzzy information and introduces a new method that can address the situations where the weight information is incompletely known and has interactive characteristics. To do this, the interval-valued intuitionistic hesitant fuzzy hybrid Shapley weighted averaging (IVIHFHSWA) operator and the interval-valued intuitionistic hesitant fuzzy hybrid Shapley weighted geometric mean (IVIHFHSWGM) operator are defined. Note that when the applied fuzzy measures are 2-additive measures, we derive the IVIHFHSWA and IVIHFHSWGM operators with respect to 2-additive measures. Using the Shapley function, models for determining the optimal fuzzy measures and 2-additive measures on the attribute set and the ordered position set are constructed. The rest of this paper is organized as follows: Section 2 first reviews several basic concepts including fuzzy measures, Choquet integral, λ-fuzzy measures and IVIHFEs. Then, the IVIHFCI operator (Joshi and Kumar, 2016) is listed. Meanwhile, it analyses the limitations of the IVIHFCI operator. Section 3 defines two new interval-valued intuitionistic hesitant fuzzy aggregation operators by using the Shapley function and briefly studies several special cases. Section 4 first recalls the hamming distance in Joshi and Kumar (2016) and points out its limitations. Then, a new hamming distance measure is defined that avoids the limitations in the Joshi and Kumar’s hamming distance. When the weight information is not exactly known, models for the optimal fuzzy measures and 2-additive measures on the criteria set and the ordered position set are constructed, respectively. Section 5 gives a new group decision making with interval-valued intuitionistic hesitant fuzzy information. Furthermore, a practical decision-making problem about the development of large projects is offered to show the concrete application of the new method. Meanwhile, the comparison analysis is made. Conclusions and future remarks are offered in the last section.
Several Concepts
This section contains three parts. The first part reviews the concepts of fuzzy measures and the Choquet integral. The second section recalls interval-valued intuitionistic hesitant fuzzy elements (IVIHFEs), several operations and a ranking order relationship. The last section lists three previous aggregation operators.
Fuzzy Measures and the Choquet Integral
Just as researchers (Beliakov and Janes, 2011; Grabisch, 1995, 1996, 1997; Fujimoto et al., 2006; Tan, 2011; Tan et al., 2011; Xu, 2010) noted, the independence of the weights of criteria in a decision-making problem is usually violated. To deal with this case, fuzzy measures introduced by Sugeno (1974) are good choices, which not only give the importance of each element but also consider the weights of all their combinations.
(See Sugeno, 1974).
A fuzzy measure on finite set X={x1,x2,…,xn} is a set function μ:P(X)→[0,1] satisfying
μ(∅)=0,μ(X)=1,
For all A,B∈P(X) with A⊆B, we have μ(A)⩽μ(B), where P(X) is the power set of X.
When X denotes the criteria set in a decision-making problem, μ(A) can be viewed as the importance of the criteria in A. Just as some researchers noted, there are three cases for the interactions between the weights of criteria. Let A and B be any two subsets in X such that A∩B=∅. When μ(A)+μ(B)=μ(A∪B), then there is no interaction between the weights of the criteria in subsets A and B. When μ(A)+μ(B)<μ(A∪B), then the complementary interaction exists between their weights. Furthermore, when μ(A)+μ(B)>μ(A∪B)\mu (A\cup B)$]]>, we know that their important interaction is redundant.
Although fuzzy measures are powerful tools to reflect the interactions between the weights of criteria, they define on the power set. This means that we need 2n−2 coefficients to determine a fuzzy measure on a set with n elements. To address this issue, some special types of fuzzy measures are introduced such as λ-fuzzy measures (Sugeno, 1974) and k-additive measures (Grabisch, 1997).
(See Sugeno, 1974).
Let X={x1,x2,…,xn} be a finite set. A fuzzy measure gλ on X is called a λ- fuzzy measure if it satisfies
gλ(A∪B)=gλ(A)+gλ(B)+λgλ(A)gλ(B),
where λ>−1-1$]]>, and A,B⊆X with A∩B=∅.
One can check that when λ=0, then gλ is an additive measure. This means that the weights of the elements in subsets A and B are independent. When λ>00$]]>, there is a complementary interaction between their weights, and the interaction between their weights is redundant for −1<λ<0.
From formula (1), one can check that the λ-fuzzy measure gλ can be equivalently expressed as:
gλ(A)=1λ(∏i∈A[1+λgλ(i)]−1)ifλ≠0,∑i∈Agλ(i)ifλ=0.
From μ(X)=1, we know that λ can be determined as:
∏i∈N[1+λgλ(i)]=1+λ.
Thus, when the importance of each element in X is known, we can use formula (3) to determine the value of λ. Then, we can further apply formula (2) to get the λ-fuzzy measure gλ.
Considering decision-making problems with interactive characteristics, it needs some other tools to obtain the comprehensive criteria values of the objects. The Choquet integral on discrete sets introduced by Grabisch (1996) is one of the most applied tools, which is explicitly defined as follows:
(See Grabisch, 1996).
Let f be a positive real-valued function on X={x1,x2,…,xn}, and μ be a fuzzy measure on X. The discrete Choquet integral of f with respect to μ is defined as:
Cμ(f(x(1)),f(x(2)),…,f(x(n)))=∑i=1nf(x(i))(μ(A(i))−μ(A(i+1))),
where (·) indicates a permutation on N such that f(x(1))⩽f(x(2))⩽⋯⩽f(x(n)), and A(i)={x(i),…,x(n)} with A(n+1)=∅.
After the pioneer work of Grabisch (1996), many decision-making methods based on the Choquet integral are developed (Beliakov, 2005; Xu, 2010; Tan, 2011; Tan et al., 2011; Joshi and Kumar, 2016).
The Concept of IVIHFEs
With the constant increasing complexity of decision making problems, many extended types of fuzzy sets are proposed. To express the uncertain membership and non-membership information as well as the hesitancy of the decision makers, Zhang (2013) presented the concept of IVIHFEs that are composed by several interval-valued intuitionistic fuzzy values (IVIFVs) (Xu and Chen, 2007).
(See Zhang, 2013).
Let X={x1,x2,…,xn} be a finite set. An interval-valued intuitionistic fuzzy set (IVIHFS) E˜ on X is defined in terms of a function that when applied to X it returns a subset of the set of all IVIFVs, denoted by E˜={⟨xi,hE˜(x)⟩,xi∈X}, where hE˜(x) is a set of several IVIFVs denoting the possible interval membership and non-membership degrees of the element xi∈X to the set E˜. For simplicity, h˜=hE˜(x) is called the interval-valued intuitionistic hesitant fuzzy element (IVIHFE), and H˜ is the set of all IVIHFEs. Any α˜∈h˜ is an IVIFV, denoted by α˜=([μl,μu],[vl,vu]).
From Definition 4, one can check that IVIHFEs can be viewed as an extension of several types of fuzzy sets including hesitant fuzzy values (HFVs) (Torra, 2010), interval-valued intuitionistic fuzzy values (IVIFVs) (Atanassov and Gargov, 1989) and interval-valued hesitant fuzzy values (IVHFVs) (Chen et al., 2013).
Following the operations on IVIFVs (Xu and Chen, 2007), Zhang (2013) defined the following operational laws on IVIHFEs. Let h˜, h˜1 and h˜2 be any three IVIHFEs in H˜. Then,
Considering the order relationship between IVIHFEs, similar to Xu and Chen (2007), Zhang (2013) introduced the concepts of the score function and the accuracy function on IVIHFEs. Let h˜ be an IVIHFE in H˜, then its score function is given as:
s(h˜)=1#h˜∑α˜∈h˜μl+μu−vl−vu2
and the accuracy function is defined as follows:
a(h˜)=1#h˜∑α˜∈h˜μl+μu+vl+vu2
where #h˜ denotes the number of IVIFVs in h˜.
Let h˜1 and h˜2 be any two IVIHFEs in H˜. Then, their order relationship is listed as follows:
If s(h˜1)<s(h˜2), then sh˜1<h˜2.
If s(h˜1)=s(h˜2), then a(h˜1)<a(h˜2)⇒h˜1<h˜2,a(h˜1)=a(h˜2)⇒h˜1=h˜2.
Several Interval-Valued Intuitionistic Hesitant Fuzzy Aggregation Operators
To compute the comprehensive values of alternatives, Zhang (2013) introduced the following two interval-valued intuitionistic hesitant fuzzy hybrid aggregation operators:
(See Zhang, 2013).
Let h˜i, i=1,2,…,n, be a collection of IVIHFEs, which has an associated weight vector w=(w1,w2,…,wn)T on the ordered set N={1,2,…,n} such that wj∈[0,1] and ∑j=1nwj=1. Then,
the interval-valued intuitionistic hesitant fuzzy hybrid averaging (IVIHFHA) operator is defined as follows:
IVIHFHAw,ω(h˜1,h˜2,…,h˜n)=⨁i=1nwih˜σ(i)′
where h˜σ(i)′ is the ith largest value of is the weighted arguments nωjh˜j, j=1,2,…,n, ω=(ω1,ω2,…,ωn)T is the weight vector on {h˜i}i∈{1,2,…,n} with ωi>00$]]> and ∑i=1nωi=1, and n is the balancing coefficient;
the interval-valued intuitionistic hesitant fuzzy hybrid geometric (IVIHFHG) operator is defined as follows:
IVIHFHGw,ω(h˜1,h˜2,…,h˜n)=⨂i=1n(h˜σ(i))wi
where h˜σ(i) is the ith largest value of is the weighted arguments h˜jnωj, j=1,2,…,n, and the other notations as shown in formula (7).
Recently, Joshi and Kumar (2016) noted that the Zhang’s aggregation operators are based on the assumption that the importance of criteria is independent. To extend the application of IVIHFEs in decision making, Joshi and Kumar (2016) used the Choquet integral with respect to the λ-fuzzy measures to define the following interval-valued intuitionistic hesitant fuzzy Choquet integral (IVIHFCI) operator: (See Joshi and Kumar, 2016).
Let μ be a fuzzy measure on X={x1,x2,…,xn}, and h˜i, i=1,2,…,n, be a collection of IVIHFEs on X. The IVIHFCI operator is a mapping IVIHFCI: H˜n→H˜, defined as:
IVIHFCI(h˜1,h˜2,…,h˜n)=⨂i=1n(h˜(i))μ(A(i))−μ(A(i+1)){[∏i=1n(μl(i))μ(A(i))−μ(A(i+1)),∏i=1n(μu(i))μ(A(i))−μ(A(i+1))],[1−∏i=1n(1−vl(i))μ(A(i))−μ(A(i+1)),1−∏i=1n(1−vu(i))μ(A(i))−μ(A(i+1))]|α˜(i)∈h˜(i),i=1,2,…,N},
where (·) is a permutation on the subscripts of the elements in X such that h˜(1)⩽h˜(2)⩽⋯⩽h˜(n), A(i)={h˜(i),…,h˜(n)} and A(n+1)=∅.
Although the IVIHFCI operator can cope with the situations where the importance of criteria in a decision-making problem is interactive, it is not a good tool to calculate the comprehensive values of the objects. As well known, the fundamental principle of the Yager’s OWA operator is to rearrange the considered elements, and their weights are only related to the ordered positions. The Choquet integral, as an extension of the OWA operator, also has this property, namely, the fuzzy measure should be defined on the ordered position set rather than on the criteria set. Otherwise, when the Choquet integral operator is applied to calculate the objects’ comprehensive values, it may endow the same criterion with the different weights just because the objects’ criteria values are different. This seems to be undesirable.
(See Joshi and Kumar, 2016).
Let us consider the decision-making problem about selecting the project manager, where the IVIHFSs E˜A1, E˜A2 and E˜A3 for the candidates A1, A2 and A3 with respect to the criteria {C1: knowledge; C2: reliability; C3: demanding } are defined as follows:
E˜A1={⟨C1,{([0.7,0.9],[0.1,0.1]),([0.6,0.8],[0.1,0.2]),([0.3,0.4].[0.6,0.6])}⟩,⟨C2{([0.5,0.6],[0.2,0.3]),([0.1,0.1],[0.8,0.9])}⟩,⟨C3,{([0.8,0.9],[0.1,0.1])}⟩},E˜A2={⟨C1,{([0.3,0.5],[0.4,0.5])}⟩,⟨C2,{([0.8,0.9],[0.1,0.1]),([0.5,0.7],[0.1,0.2])}⟩,⟨C3,{([0.2,0.3],[0.5,0.6]),([0.1,0.3],[0.6,0.6])}⟩}
and
E˜A3={⟨C1,{([0.2,0.4],[0.3,0.5]),([0.5,0.7],[0.1,0.2])}⟩,⟨C2,{([0.1,0.1],[0.7,0.9])}⟩,⟨C3,{([0.1,0.3],[0.6,0.7]),([0.2,0.2],[0.7,0.8]),([0.3,0.4],[0.6,0.6])}⟩}.
According to the ranking method on IVIHFEs, we derive h˜2(A1)<h˜1(A1)<h˜3(A1), h˜3(A3)<h˜1(A3)<h˜2(A3) and h˜2(A4)<h˜3(A4)<h˜1(A4) with respect to E˜A1, E˜A2 and E˜A3, respectively. When μ(C1)=μ(C3)=0.4 and μ(C2)=0.3, using formula (3) we derive λ=−0.258. Thus, the fuzzy measures are
μ(C1,C2)=μ(C2,C3)=0.669,μ(C1,C3)=0.769,μ(C1,C2,C3)=1.
From the IVIHFCI operator, we know that the weights of the criteria C1, C2, and C3 are 0.369, 0.231 and 0.4 for the candidate A1, their weights are respective of 0.369, 0.3 and 0.331 for the candidate A2, while they are 0.4, 0.231 and 0.369 for the candidate A3 (see Table 1). None of them are the same for these candidates. This means that the evaluations of these three candidates are based on the different standards. It is unreasonable to give their ranking orders by using the comprehensive IVIHFEs obtained from the IVIHFCI operator.
Although the λ-fuzzy measures can simplify the complexity of determining fuzzy measures, especially, when the set has a large number of elements, there are some limitations. From formula (3), one can verify that −1<λ<0, when the sum of each element’s weight is bigger than one; If the sum of each element’s weight is smaller than one, we derive λ>00$]]>, and we have λ=0 when the sum of each element’s weight is equal to one. This shows that their interactions are completely determined by their respective weights.
The weights of the criteria with respect to the different candidates.
The candidates
The weights of the criteria
C1
C2
C3
A1
0.369
0.231
0.400
A2
0.369
0.300
0.331
A3
0.400
0.231
0.369
Furthermore, the λ-fuzzy measures can only reflect the complementary interactions, redundant interactions or independency between the elements’ weights. The question is when the exact fuzzy measure on an element set is unknown, how can we ensure that there are only complementary interactions, redundant interactions or independency between their weights? Furthermore, the interactive characteristics should be determined by the characteristics of criteria themselves rather than their respective weights.
These properties indicate that the λ-fuzzy measures are unsuitable to give the interactions between the weights of criteria. For example, Grabisch (1995) introduced the well-known example to show the application of the Choquet integral, which is about the evaluation of the students in relation to three subjects: {mathematics, physics, literature}, where the importance is respectively defined by 0.375, 0.375 and 0.3. On the other hand, more importance is attributed to science-related subjects than to literature, but some advantages are given to the students that are good both in literature and in any of the science-related subjects. In this situation, the λ-fuzzy measures seem to be helpless since λ=0 by formula (3), which cannot reflect interactions between the weights of these three subjects.
In this example, one can find that the interactions between the weights of the criteria have no direct relationship with respect to the sum of each criterion’s weight. Furthermore, there may be complementary and redundant interactions simultaneously.
New Interval-Valued Intuitionistic Hesitant Fuzzy Aggregation Operators
Considering the previous aggregation operators that cannot well address the situation where the weights of elements in a set are interactive, the section defines several new ones. Before defining the new aggregation operators, let us first introduce the following two new operations on IVIHFEs:
h˜1λ1⊗h˜2λ2={α˜iλ1⊗α˜jλ2|α˜i∈h˜1,α˜j∈h˜2}={([(μli)λ1(μlj)λ2,(μui)λ1(μuj)λ2],[(vli)λ1(vlj)λ2,(vui)λ1(vuj)λ2])|α˜i∈h˜1,α˜j∈h˜2}, where λ1,λ2∈[0,1] with λ1+λ2⩽1.
The above new operations can avoid some limitations that are listed in Section 2.2. Let h˜1 and h˜2 be any two IVIHFEs, if we have h˜1<h˜2 according to formulae (5) and (6). Then, we might derive h˜1⊕h˜>h˜2⊕h˜h˜1⊗h˜>h˜2⊗h˜{\tilde{h}_{2}}\oplus \tilde{h}\\ {} {\tilde{h}_{1}}\otimes \tilde{h}>{\tilde{h}_{2}}\otimes \tilde{h}\end{array}\right.$]]> for some IVIHFE h˜ and λh˜1>λh˜2h˜1λ>h˜2λ\lambda {\tilde{h}_{2}}\\ {} {\tilde{h}_{1}^{\lambda }}>{\tilde{h}_{2}^{\lambda }}\end{array}\right.$]]> for λ∈[0,1]. For example, let h˜1={[0.3,0.4],[0.4,0.6],[0.4,0.6],[0.3,0.4]}, h˜2={([0.2,0.3],[0.2,0.3],([0.4,0.5],[0.3,0.5])} and h˜={([0.6,0.7],[0.1,0.2])}. According to formula (5), we derive h˜1<h˜2 for s(h˜1)=0 and s(h˜1)=0.05. However, we derive h˜1⊕h˜>h˜2⊕h˜{\tilde{h}_{2}}\oplus \tilde{h}$]]> according to the first operational law listed in Section 2.2, where s(h˜1⊕h˜)=0.7275>s(h˜2⊕h˜)=0.7125s({\tilde{h}_{2}}\oplus \tilde{h})=0.7125$]]> with h˜1⊕h˜={([0.72,0.82],[0.04,0.12]),([0.76,0.88],[0.03,0.08])} and h˜2⊕h˜={([0.68,0.79],[0.02,0.06])([0.76,0.85],[0.03,0.01])}. Furthermore, let λ=0.3, we get h˜1λ>h˜2λ{\tilde{h}_{2}^{\lambda }}$]]> using the fourth operational law listed in Section 2.2, where s(h˜1λ)=0.6120>s(h˜2λ)=0.6076s({\tilde{h}_{2}^{\lambda }})=0.6076$]]> with h˜1λ={([0.69,0.76],[0.14,0.24]),([0.76,0.86],[0.10,0.14])} and h˜2λ={([0.62,0.69],[0.06,0.10]),([0.76,0.81],[0.10,0.19])}. For the other two operations, one can similarly derive the above conclusions.
Without loss of generality, all of the following operations use the operational laws offered in Definition 7.
From the analysis in Section 2.3, we know that it is unreasonable to calculate the compared objects’ comprehensive IVIHFEs by using the IVIHFCI operator. To reflect the interactions between the weights of the criteria as well as to ensure the weights of criteria are the same for all objects, we apply the Shapley function in game theory (Shapley, 1953) to define two hybrid aggregation operators on IVIHFEs.
The Shapley function (Shapley, 1953) is one of the most important payoff indices in cooperative game theory, which satisfies many desirable properties. When the Shapley function is restricted in the setting of fuzzy measures, we derive:
Shxi(μ,X)=∑S⊆X∖xi(n−s−1)!s!n!(μ(S∪xi)−μ(S)),∀xi∈X,
where μ is a fuzzy measure on X={x1,x2,…,xn}, n and s denote the numbers of the elements in X and S, respectively.
From formula (10), we know that the Shapley function gives the global interactions between each element xi and all coalitions in Xi. When X denotes the criteria set in a decision-making problem, μ(S∪xi)−μ(S) can be viewed as the contribution of the xi’s importance to the importance of the coalition S.
Let μ be a fuzzy measure on X={x1,x2,…,xn}, and Sh be the Shapley function as shown in formula (10).
Let xi and xj be two elements in X. If we have μ(S∪xi)=μ(S∪xj) for all S⊆X∖{xi,xj}, then Shxi(μ,X)=Shxj(μ,X);
Let xi be an element in X. If we have μ(S∪xi)=μ(S) for all S⊆X∖xi, then Shxi(μ,X)=0;
Let xi be an element in X. If we have μ(S∪xi)=μ(S)+μ(xi) for all S⊆X∖xi, then Shxi(μ,X)=μ(xi);
The Shapley value vector {Shxi(μ,X)}xi∈X is a normalized weight vector on X, namely, Shxi(μ,X)⩾0 for all xi∈X and ∑xi∈XShxi(μ,X)=1.
From formula (10), one can easily derive the conclusions listed in Property 1. □
The first conclusion shows that when two elements have the same contribution to all coalitions in X∖{xi,xj}, then they have the same importance. Especially, when μ is an additive measure, we derive μ(xi)=μ(xj). The second conclusion shows that when the element xi has no contribution to the importance of all coalitions in X∖xi, then its importance is zero. The third conclusion indicates that when the contribution of the element xi to the importance of all coalitions in X∖xi is μ(xi), then its importance equals to itself. The last conclusion shows that the vector composed by the elements’ Shapley values is a weight vector. All these conclusions can be seen as a natural extension of additive measures.
Now, let us define the following interval-valued intuitionistic hesitant fuzzy Shapley weighted averaging (IVIHFSWA) operator and interval-valued intuitionistic hesitant fuzzy Shapley weighted geometric mean (IVIHFSWGM) operator.
Let μ be a fuzzy measure on X={x1,x2,…,xn}, and h˜i, i=1,2,…,n, be a collection of IVIHFEs for X.
The IVIHFSWA operator is defined as follows:
IVIHFSWA(h˜1,h˜2,…,h˜n)=⨁i=1n(Shxi(μ,X)h˜i={[∑i=1nShxi(μ,X)μli,∑i=1nShxi(μ,X)μui],[∑i=1nShxi(μ,X)vli,∑i=1nShxi(μ,X)vui]|α˜i∈h˜i,ı=1,2,…,n};
The IVIHFSWGM operator is defined as follows:
IVIHFSWGM(h˜1,h˜2,…,h˜n)=⨂i=1n(h˜i)Shxi(μ,X)={[∏i=1n(μli)Shxi(μ,X),∏i=1n(μui)Shxi(μ,X)],[∏i=1n(vli)Shxi(μ,X),∏i=1n(vui)Shxi(μ,X)]|α˜i∈h˜i,i=1,2,…,n},
where Shxi(μ,X) is the Shapley value of the element xi with respect to the fuzzy measure μ.
One easily shows that all properties for the IVIHFCI operator still hold for the IVIHFSWA and IVIHFSWGM operators.
In Example 1, when we apply the criteria’s Shapley values as their weights, we derive
ShC1(μ,C)=ShC3(μ,C)=0.367,ShC2(μ,C)=0.267,
which addresses the issue in the IVIHFCI operator.
Just as some researchers (Lin and Jiang, 2014; Merigo and Casanovas, 2009; Xu and Da, 2002; Xu, 2004) noted, the IVIHFSWA and IVIHFSWGM operators only consider the importance of the elements, but the importance of the ordered positions is not enclosed. To deal with this issue, we define the interval-valued intuitionistic hesitant fuzzy hybrid Shapley weighted averaging (IVIHFHSWA) operator and the interval-valued intuitionistic hesitant fuzzy hybrid Shapley weighted geometric mean (IVIHFHSWGM) operator as follows:
Let μ be a fuzzy measure on X={x1,x2,…,xn}, let v be a fuzzy measure on the ordered position set N={1,2,…,n}, and let h˜i(i=1,2,…,n) be a collection of IVIHFEs on X.
The IVIHFHSWA operator is defined as follows:
IVIHFHSWA(h˜1,h˜2,…,h˜n)=∑j=1nShj(v,N)Shx(j)(μ,X)h˜(j)∑j=1nShj(v,N)Shx(j)(μ,X)={[∑j=1nShj(v,N)Shx(j)(μ,X)μl(j)∑j=1nShj(v,N)Shx(j)(μ,X),∑j=1nShj(v,N)Shx(j)(μ,X)μu(j)∑j=1nShj(v,N)Shx(j)(μ,X)],[∑j=1nShj(v,N)Shx(j)(μ,X)vl(j)∑j=1nShj(v,N)Shx(j)(μ,X),∑j=1nShj(v,N)Shx(j)(μ,X)vu(j)∑j=1nShj(v,N)Shx(j)(μ,X)]|α˜i∈h˜i,ı=1,2,…,n};
The IVIHFHSWGM operator is defined as follows:
IVIHFHSWA(h˜1,h˜2,…,h˜n)=⨂i=1n(h˜(j))Shj(v,N)Shx(j)(μ,X)∑j=1nShj(v,N)Shx(j)(μ,X)={[∏j=1n(μl(j))Shj(v,N)Shx(j)(μ,X)∑j=1nShj(v,N)Shx(j)(μ,X),∏j=1n(μu(j))Shj(v,N)Shx(j)(μ,X)∑j=1nShj(v,N)Shx(j)(μ,X)],∏j=1n(vl(j))Shj(v,N)Shx(j)(μ,X)∑j=1nShj(v,N)Shx(j)(μ,X),∏j=1n(vu(j))Shj(v,N)Shx(j)(μ,X)∑j=1nShj(v,N)Shx(j)(μ,X)|α˜i∈h˜i,i=1,2,…,n},
where (·) is a permutation on Shxi(μ,X)h˜i, i=1,2,…,n, such that Shx(j)(μ,X)h˜(j) is the jth smallest value of Shxi(μ,X)h˜i, i=1,2,…,n for the IVIHFHSWA operator, and (·) is a permutation on h˜iShxi(μ,X), i=1,2,…,n, such that h˜(j)Shx(j)(μ,X) is the j th smallest value of h˜iShxi(μ,X), i=1,2,…,n, for the IVIHFHSWGM operator, Shj(v,N) is the Shapley value of the jth ordered position with respect to the fuzzy measure v, and Shxi(μ,X) is the Shapley value of the element xi for the fuzzy measure G.
The IVIHFHSWA and IVIHFHSWGM operators not only consider the importance of the elements and the ordered positions, but also reflect their respective interactions. Furthermore, these two operators can be seen as extensions of many interval-valued intuitionistic hesitant fuzzy aggregation operators.
(i) When there are no interactions between the weights of the ordered positions in N as well as between the weights of the elements in X, then the IVIHFHSWA operator degenerates to the interval-valued intuitionistic hesitant fuzzy hybrid weighted averaging (IVIHFHWA) operator
IVIHFHWA(h˜1,h˜2,…,h˜n)=∑j=1nwjωx(j)h˜(j)∑j=1nwjωx(j)={[∑j=1nwjωx(j)μl(j)∑j=1nwjωx(j),∑j=1nwjωx(j)μu(j)∑j=1nwjωx(j)],[∑j=1nwjωx(j)vl(j)∑j=1nwjωx(j),∑j=1nwjωx(j)vu(j)∑j=1nwjωx(j)]α˜i∈h˜i,i=1,2,…,n},
and the IVIHFHSWGM operator degenerates to the interval-valued intuitionistic hesitant fuzzy hybrid weighted geometric mean (IVIHFHWGM) operator
IVIHFHWGM(h˜1,h˜2,…,h˜n)=⨂i=1n(h˜(j))wjωx(j)∑j=1nwjωx(j)={[∏j=1n(μl(j))wjωx(j)∑j=1nwjωx(j),(μu(j))wjωx(j)∑j=1nwjωx(j)],[(vl(j))wjωx(j)∑j=1nwjωx(j),(vu(j))wjωx(j)∑j=1nwjωx(j)]|α˜i∈h˜i,i=1,2,…,n},
where ω=(ωx1,ωx2,…,ωxn) is an additive weight vector on X, w=(w1,w2,…,wn) is an additive weight vector on N, (·) is a permutation on ωxih˜i, i=1,2,…,n, such that ωx(j)h˜(j) is the jth smallest value of ωxih˜i, i=1,2,…,n, for the IVIHFHWA operator, and (·) is a permutation on h˜iωxi, i=1,2,…,n, such that h˜(j)ωx(j) is the jth smallest value of h˜iωxi, i=1,2,…,n, for the IVIHFHWGM operator.
(ii) When there are no interactions between the weights of the ordered positions in N and w=(1n,1n,…,1n), then the IVIHFHSWA operator degenerates to the IVIHFSWA operator, and the IVIHFHSWGM operator degenerates to the IVIHFSWGM operator.
(iii) When there are no interactions between the weights of the elements in X and ω=(1n,1n,…,1n), then the IVIHFHSWA operator degenerates to the interval-valued intuitionistic hesitant fuzzy ordered Shapley weighted averaging (IVIHFOSWA) operator
IVIHFOSWA(h˜1,h˜2,…,h˜n)=∑j=1nShj(v,N)h˜(j)={[∑j=1nShj(v,N)μl(j),∑j=1nShj(v,N)μu(j)],[∑j=1nShj(v,N)vl(j),∑j=1nShj(v,N)vu(j)]|α˜i∈h˜i,i=1,2,…,n},
and the IVIHFHSWGM operator degenerates to the interval-valued intuitionistic hesitant fuzzy ordered Shapley weighted geometric mean (IVIHFOSWGM) operator
IVIHFOSWGM(h˜1,h˜2,…,h˜n)=⨂i=1nh˜(j)Shj(v,N)={[∏j=1n(μl(j))Shj(v,N),∏j=1n(μu(j))Shj(v,N)],[∏j=1n(vl(j))Shj(v,N),∏j=1n(vu(j))Shj(v,N)]|α˜i∈h˜i,i=1,2,…,n},
where (·) is a permutation on the IVIHFEs h˜i, i=1,2,…,n, such that h˜(j) is the jth smallest value of h˜i, i=1,2,…,n.
Furthermore, when there are no interactions between the weights of the ordered positions in N, then the IVIHFOSWA operator reduces to the interval-valued intuitionistic hesitant fuzzy ordered weighted averaging (IVIHFOWA) operator
IVIHFOWA(h˜1,h˜2,…,h˜n)=∑j=1nwjh˜(j)={[∑j=1nwjμl(j),∑j=1nwjμu(j)],[∑j=1nwjvl(j),∑j=1nwjvu(j)]|α˜i∈h˜i,i=1,2,…,n},
and the IVIHFOSWGM operator reduces to the interval-valued intuitionistic hesitant fuzzy ordered weighted geometric mean (IVIHFOWGM) operator
IVIHFOWGM(h˜1,h˜2,…,h˜n)=⨂i=1nh˜(j)wj={[∏j=1n(μl(j))wj,∏j=1n(μu(j))wj],[∏j=1n(vl(j))wj,∏j=1n(vu(j))wj]|α˜i∈h˜i,i=1,2,…,n},
where w=(w1,w2,…,wn) is an additive weight vector on N, and (·) is a permutation on the IVIHFEs h˜i, i=1,2,…,n, such that h˜(j) is the jth smallest value of h˜i, i=1,2,…,n.
To overcome the limitations of the λ-fuzzy measures, 2-additive measures introduced by Grabisch (1997) are good choices to reduce the complexity of determining fuzzy measures.
(See Grabisch, 1997).
A fuzzy measure μ on N={1,2,…,n} is said to be a 2-additive measure, if, for any S⊆N with s⩾2, we have
μ(S)=∑{i,j}⊆Sμ(i,j)−(s−2)∑i∈Sμ(i)
where s is the cardinality of S.
From the concept of 2-additive measures, it only needs n(n−1)/2 coefficients to determine a fuzzy measure on a set with n elements. Furthermore, we have the following conclusion.
(See Grabisch, 1997).
Let μ be a fuzzy measure onN={1,2,…,n}, then μ is a 2-additive measure if and only if there exist coefficientsμ(i)andμ(i,j)for alli,j∈Nthat satisfy the following conditions:
μ(i)⩾0,i∈N,
∑{i,j}⊆Nμ(i,j)−(n−2)∑i∈Nμ(i)=1,
∑i⊆S∖kμ(i,k)−μ(i)⩾(s−2)μ(k)∀S∈N s.t. k∈Swiths⩾2,
where s and n denote the cardinalities of S and N, respectively.
When μ is a 2-additive measure, Meng and Tang (2013) gave the following conclusion: (See Meng and Tang, 2013).
Let μ be a 2-additive measure defined onN={1,2,…,n}, then the Shapley functionShcan be expressed as follows:Shi(μ,N)=3−n2μ(i)+12∑j∈N∖iμ(i,j)−μ(j),∀i∈N.
From formula (16), we know that n(n+1)/2 coefficients are needed to determine a 2-additive measure on a set with n elements. Although to determine a 2-additive measure demands more coefficients than to derive a λ-fuzzy measure, 2-additive measures can reflect the complementary interactions, redundant interactions and independency between the weights of the elements simultaneously. Furthermore, their interactive characteristics have no relationship to the sum of each criterion’s weight in the setting of 2-additive measures. Thus, it is more reasonable to use 2-additive measures than to apply λ-fuzzy measures.
When the fuzzy measure μ and the fuzzy measure v are both a 2-additive measure, then we derive the associated IVIHFHSWA and IVIHFHSWGM operators with respect to 2-additive measures. It is worth noting that both the IVIHFSWG operator and the IVIHFHSWG operator overcome the issues in the IVIHFCI operator.
Hamming Distance Based Models for the Optimal Fuzzy Measures
This section focuses on how to determine the optimal fuzzy measures when the weight information is partly known. Similarly, the 2-additive measures can also be derived by using the built models.
A New Hamming Distance Measure on IVIHFEs
Joshi and Kumar (2016) defined the hamming distance on IVIHFEs to rank objects, but this hamming distance requires the calculated IVIHFEs to have the same length; otherwise, we need to add some IVIFVs in the shorter IVIHFE several times until its length equals to the longer one.
Let h˜1 and h˜2 be two IVIHFEs, where
h˜1={α˜1=([0.3,0.4],[0.1,0.2]),α˜2=([0.5,0.6],[0.2,0.3]),α˜3=([0.6,0.7],[0.1,0.2])}
and
h˜2={β˜1=([0.1,0.2],[0.4,0.5]),β˜2=([0.3,0.4],[0.1,0.2])}.
According to Definition 2.5 in Joshi and Kumar (2016), we need to add one IVIFV into h˜2. When we add the first IVIFV in h˜2, we derive h˜2′={β˜1,β˜1,β˜2}. Then, we can apply formula (8) in (Joshi and Kumar, 2016) to calculate the hamming distance between h˜1 and h˜2, which is in fact the hamming distance between h˜1 and h˜2′. However, one can check that h˜2′ and h˜2 are not equivalent. Using formula (5), we obtain their scores s(h˜2)=−0.05 and s(h˜2′)=−0.13, by which we get h˜2>h˜2′{\tilde{h}^{\prime }_{2}}$]]>. On the other hand, when we add the second IVIFV in h˜2, we have h˜2″={β˜1,β˜2,β˜2}. Using formula (5), we obtain its score s(h˜2″)=−0.03, by which we get h˜2<h˜2″. Furthermore, according to Definition 2.5, the hamming distance between h˜1 and h˜2 is
d(h˜1,h˜2)=13×4(|α˜1−β˜1|+|α˜2−β˜1|+|α˜3−β˜2|)=0.23
for h˜2′, and it is
d(h˜1,h˜2)=13×4(|α˜1−β˜1|+|α˜2−β˜2|+|α˜3−β˜2|)=0.18
for h˜2″, where |α˜i−β˜j|=|μli−μlj|+|μui−μuj|+|vli−vlj|+|vui−vuj| with i=1,2,3; j=1,2. Thus, it is unreasonable to add IVIFVs into IVIHFEs subjectively.
For α˜1=([0.3,0.4],[0.1,0.2]), we know |α˜1−β˜1|=1 according to formula (8) in Joshi and Kumar (2016). However, we have |α˜1−β˜2|=0 for α˜1=β˜2. In this case, we should choose the latter for α˜1∈h˜2. This also shows that it is undesirable to use formula (8) in Joshi and Kumar (2016) to calculate the hamming distance between IVIHFEs.
Next, we introduce a new hamming distance measure on IVIHFEs, which needn’t consider the length of IVIHFEs and the arrangement of their possible IVIFVs.
Let h˜1 and h˜2 be two IVIHFEs. Without loss of generality, suppose that h˜1={α˜i}i=1,2,…,m and h˜2={β˜j}j=1,2,…,n, where α˜i=([μli,μui],[vli,vui]) and β˜j=([ηlj,ηuj],[κlj,κuj]) for all i=1,2,…,m;j=1,2,…,n. Then, the distance measure from α˜i to h˜2 is defined as follows:
D(α˜i,h˜2)→=minβ˜j∈h˜2|μli−ηlj|+|μui−ηuj|+|vli−κlj|+|vui−κuj|4.
Let h˜1 and h˜2 be two IVIHFEs, where h˜1={α˜i}i=1,2,…,m and h˜2={β˜j}j=1,2,…,n. Then, the hamming distance measure between h˜1 and h˜2 is defined as:
D(h˜1,h˜2)=D(h˜1,h˜2)→+D(h˜2,h˜1)→2,
where D(h˜1,h˜2)→=14m∑i=1mD(α˜i,h˜2)→ and D(h˜2,h˜1)→=14n∑j=1nD(β˜j,h˜1)→.
Let h˜1 and h˜2 be any two IVIHFEs, where h˜1={α˜i}i=1,2,…,m and h˜2={β˜j}j=1,2,…,n. Then,
D(h˜1,h˜2)=0 if and only if h˜1=h˜2, namely, there exists β˜j such that α˜i=β˜j for all α˜i∈h˜1, and there is α˜i such that β˜j=α˜i for all β˜j∈h˜2;
D(h˜1,h˜2)=1 if and only if h˜1={([1,1],[0,0])} and h˜2={([0,0],[1,1])} or h˜1={([0,0],[1,1])} and h˜1={([1,1],[0,0])}. Otherwise, we have 0<D(h˜1,h˜2)<1 with h˜1≠h˜2;
D(h˜1,h˜2)=D(h˜1,h˜2).
From formulae (17) and (18), it is not difficult to derive the results in Property 2. □
In Example 2, when the new hamming distance measure on IVIHFEs is applied, we have
D(h˜1,h˜2)→=112(|α˜1−β˜2|+|α˜2−β˜2|+|α˜3−β˜2|)=112(0+0.6+0.6)=0.1
and
D(h˜1,h˜2)→=18(|β˜1−α˜1|+|β˜2−α˜1|)=18(1+0)=0.125,
by which the hamming distance between h˜1 and h˜2 is D(h˜1,h˜2)=0.1125. It is different to the values obtained from Joshi and Kumar’s hamming distance.
From the example given in Joshi and Kumar (2016), when we use the IVIHFSWA and IVIHFSWGM operators as well as the new hamming distance measure, the ranking order is A1>A3>A2>A4{A_{3}}>{A_{2}}>{A_{4}}$]]>, which is different from that obtained by using the Joshi and Kumar’s method.
Models for the Optimal Fuzzy Measures and 2-Additive Measures
Considering a multi-criteria decision making problem, let A={A1,A2,…,Am} be the set of alternatives, which are evaluated by a decision maker team according to the criteria set C={C1,C2,…,Cn} by using IVIFVs. When the decision makers have different opinions for the alternatives’ criteria values, there can be several IVIFVs to denote the criteria values, which are denoted using IVIHFEs. Suppose that the evaluation of the alternative Ai with respect to the attribute Cj is an IVIHFE h˜ij(i=1,2,…,m;j=1,2,…,n). By H˜=(h˜ij)m×n, we denote the IVIHFE matrix given by the decision maker team.
When all criteria Cj, j=1,2,…,n, are benefit (i.e. the larger the greater preference), then the criteria values needn’t normalization; otherwise, we normalize the IVIHFE matrix H˜=(h˜ij)m×n into H˜′=(h˜ij′)m×n, where h˜ij′=h˜ijfor benefit criteriaCj(h˜ij)cfor cost criteriaCj with (h˜ij)c={(α˜ijk)c=([vlijk,vuijk],[μlijk,μuijk])|α˜ijk=([μlijk,μuijk],[vlijk,vuijk])∈h˜ij,k=1,2,…,h˜ij} for all i=1,2,…,m; j=1,2,…,n.
When the fuzzy measure μ on the criteria set C and the fuzzy measure v on the ordered position set N are exactly known, we can adopt the aggregation operator given in Section 3 to calculate the collective values. However, when we do not own the fully weight information, it needs to first determine the weight vectors on them. Next, we construct Hamming distance measure based models to determine the optimal fuzzy measures and 2-additive measures on the criteria set C and the ordered position set N, which can be seen as the extensions of the additive measures.
Because we cannot guarantee the importance of the criteria is independent, we apply the Shapley function to give the weights of the criteria. With respect to the normalized IVIHFE matrix H˜′=(h˜ij′)m×n, let
h˜j′+=max1⩽i⩽m{max1⩽k⩽#h˜ijα˜ij′k|α˜ij′k∈h˜ij′}
and
h˜j′−=min1⩽i⩽m{min1⩽k⩽#h˜ijα˜ij′k|α˜ij′k∈h˜ij′}
for each j=1,2,…,n.
Because the weight information makes the criteria values the bigger the better, when the weight information on the criteria set C is not exactly known, we construct the following model to determine the optimal fuzzy measure μ.
min∑i=1m∑j=1nD(h˜ij′,h˜j′+)D(h˜ij′,h˜j′+)+D(h˜ij′,h˜j′−)ShCj(μ,C)s.t.μ(S)⩽μ(T),∀S⊆T⊆C,μ(Ci)∈WCi,i=1,2,…,n,μ(Ci)⩾0,i=1,2,…,n,
where WCi is the known weight information of the criterion Cj, j=1,2,…,n.
When the optimal fuzzy measure μ is a 2-additive measure, model (19) can be equivalently transformed into the following model:
min∑i=1m∑j=1nD(h˜ij′,h˜j′+)D(h˜ij′,h˜j′+)+D(h˜ij′,h˜j′−)ShCj(μ,C)s.t.∑Cj∈S∖Ciμ(Ci,Cj)−μ(Cj)⩾(s−2)μ(Ci),∀S⊆C,∀Ci∈S,s⩾2,∑{Ci,Cj}⊆Cμ(Ci,Cj)−(n−2)∑Ci∈Cμ(Ci)=1,μ(Ci)∈WCi,i=1,2,…,n,μ(Ci)⩾0,i=1,2,…,n,
where the notations as listed in model (19).
From formula (16), we further have the following model for the optimal 2-additive measure μ:
min3−n2∑i=1m∑j=1nD(h˜ij′,h˜j′+)D(h˜ij′,h˜j′+)+D(h˜ij′,h˜j′−)μ(Cj)+12∑i=1m∑j=1n∑k=1,k≠jnD(h˜ij′,h˜j′+)D(h˜ij′,h˜j′+)+D(h˜ij′,h˜j′−)(μ(Cj,Ck)−μ(Ck))s.t.∑Cj∈S∖Ciμ(Ci,Cj)−μ(Cj)⩾(s−2)μ(Ci),∀S⊆C,∀Ci∈S,s⩾2,∑{Ci,Cj}⊆Cμ(Ci,Cj)−(n−2)∑Ci∈Cμ(Ci)=1,μ(Ci)∈WCi,i=1,2,…,n,μ(Ci)⩾0,i=1,2,…,n.
To reduce the influence of the unduly high or low evaluation values induced by the decision makers’ limited knowledge or decision expertise, when the weight information on the ordered position set N is incompletely known, we construct the following model to determine the optimal fuzzy measure v on the ordered position set N:
min∑i=1m(∑j=1d(n)(1−Ri(j))Shj(μ,N)+∑j=d(n)+1nRi(j)Shj(μ,N))s.t.μ(S)⩽μ(T),∀S⊆T⊆N,μ(i)∈Wi,i=1,2,…,n,μ(i)⩾0,i=1,2,…,n,
where Ri(j)=D(h˜i(j),h˜(j)+)D(h˜i(j),h˜(j)+)+D(h˜i(j),h˜(j)−) is the jth smallest value for Rij=D(h˜ij,h˜j+)D(h˜ij,h˜j+)+D(h˜ij,h˜j−), j=1,2,…,n, for each i=1,2,…,m, mid(n)=n2nis an even number,n+12nis an odd number, and Wi is the known weight information of the jth ordered position, j=1,2,…,n.
When the optimal fuzzy measure v is a 2-additive measure, model (22) can be equivalently transformed into the following model:
min3−n2∑i=1m(∑j=1d(n)1−Ri(j)v(j)+∑j=mid(n)+1nRi(j)v(j))+12∑i=1m(∑j=1mid(n)∑k=1,k≠jn(1−Ri(j))(v(j,k)−μ(k))+∑j=d(n)+1n∑k=1,k≠jnRi(j)(v(j,k)−μ(k)))s.t.∑j∈N∖i(v(i,j)−v(j))⩾(s−2)v(i),∀S⊆N,∀i∈S,s⩾2,∑{i,j}∈Nv(i,j)−(n−2)∑i∈Nv(i)=1,v(i)∈Wi,i=1,2,…,n,v(i)⩾0,i=1,2,…,n,
where the notations as listed in model (22).
When we further require the 2-additive measure v on the ordered position set N to be symmetric for the middle position mid(n), then we establish the following model for the optimal symmetric 2-additive measure v:
min3−n2∑i=1m(∑j=1mid(n)(1−Ri(j))v(j)+∑j=mid(n)+1nRi(j)v(j))+12∑i=1m(∑j=1mid(n)∑k=1,k≠jn(1−Ri(j))(v(j,k)−μ(k))+∑j=mid(n)+1n∑k=1,k≠jnRi(j)(v(j,k)−μ(k)))s.t.∑j∈N∖iv(i,j)−v(j)⩾(s−2)v(i),∀S⊆N,∀i∈S,s⩾2,∑{i,j}⊆Nv(i,j)−(n−2)∑i∈Nv(i)=1,v(i)=v(n−i+1),∀i∈N,v(i,j)=v(n−i+1,n−j+1),∀i,j∈N,v(i)∈Wi,i=1,2,…,n,v(i)⩾0,i=1,2,…,n,
where the notations as shown in model (22).
Note that when there is no interaction between the weights of the ordered positions and that of criteria, models (19) and (22) reduce to the optimal additive measures on them, respectively.
A New Approach to Decision Making with IVIHFEs
Because of the complexity of decision-making problems, it is difficult or even impossible to require a decision maker to consider all aspects of a decision-making problem. Thus, group decision making attracts considerable attention from researchers (Beliakov et al., 2014; Liu et al., 2017b; Meng et al., 2017a, 2017b, 2017c; Pérez et al., 2010, 2014; Perez et al., 2016; Wu et al., 2017).
Based on the new defined aggregation operators with respect to 2-additive measures and the built programming models, this section gives a new group decision-making method with interval-valued intuitionistic hesitant fuzzy information.
With respect to the decision-making problem listed in Section 4.2, the following procedure is needed to rank objects:
Step 1: Let H˜′=(h˜ij′)m×n be the normalized IVIHFE matrix of H˜=(h˜ij)m×n, let μ be the fuzzy measure on the criteria set C, and let v be the fuzzy measure on the ordered position set N={1,2,…,n};
Step 2: When the fuzzy measures μ and v are not exactly known, we adopt models (21) and (24) to determine the associated optimal 2-additive measures, otherwise, go to the next step;
Step 3: We use the IVIHFHSWA or IVIHFHSWGM operator to calculate the alternatives’ comprehensive IVIHFEs h˜i, i=1,2,…,m;
Step 4: With respect to the comprehensive IVIHFEs h˜i, i=1,2,…,m, we apply the score function in Zhang (2013) to rank objects Ai, i=1,2,…,m, where s(h˜i)=12∗#h˜i∑α˜=([μl,μu],[vl,vu])∈h˜i(μl+μu−vl−vu) with #h˜i being the number of IVIFVs in h˜i, i=1,2,…,m.
To show the concrete application of the above algorithm, we offer the following decision-making problem about the development of large projects.
The enterprise’s board of directors is to plan the development of large projects strategy initiatives for the following five years. There are four possible projects Ai, i=1,2,3,4, to be evaluated. It is necessary to compare these projects to select the most important one as well as order them from the point of view of their importance, taking into account four attributes suggested by the Balanced Scorecard methodology (it should be noted that all of them are of the maximization type): C1: financial perspective, C2: the customer satisfaction, C3: internal business process perspective, and C4: learning and growth perspective. To avoid influencing each other, the decision makers are required to provide their preferences in anonymity and the IVIHFE matrix H˜=(h˜ij)4×4 is presented in Table 2, where h˜ij, i,j=1,2,3,4, is in the form of IVIHFEs.
Assume that the attribute weights are defined by WC1=[0.1,0.3], WC2=[0.2,0.4], WC3=[0.05,0.25] and WC3=[0.25,0.45], respectively. Furthermore, the weights on the ordered positions are given as follows: W1=[0.1,0.3], W2=[0.2,0.4], W3=[0.2,0.4] and W4=[0.1,0.3].
To rank these four possible projects, the following procedure is needed:
Step 1: Using model (21), the optimal 2-additive measure μ on the criteria set C can be derived, by which the Shapley values are ShC1(μ,C)=0.025,ShC2(μ,C)=0.400, ShC3(μ,C)=0.475, ShC4(μ,C)=0.100.
Step 2: Using model (24), we can obtain the optimal symmetric 2-additive measure v on the ordered set N and the associated Shapley values, where Sh1(v,N)=Sh2(v,N)=Sh3(v,N)=Sh4(v,N)=0.25.
Step 3: Adopting the IVIHFHSWA operator, we can derive the comprehensive IVIHFEs of the projects. Taking the collective IVIHFE of the project a1, for example, we have:
h˜1={([0.3450,0.5425],[0.3125,0.4125]),([0.3500,0.5463],[0.3100,0.4100]),([0.4650,0.6625],[0.2325,0.3325]),([0.4700,0.6663],[0.2300,0.3300]),([0.3750,0.5625],[0.2925,0.3925]),([0.3800,0.5663],[0.2900,0.3900]),([0.4950,0.6825],[0.2125,0.3125]),([0.5000,0.6863],[0.2100,0.3100])}.Step 4: With respect to the comprehensive IVIHFEs h˜i, i=1,2,3,4, the scores are derived as follows:
s(h˜1)=0.2072,s(h˜2)=0.1075,s(h˜3)=0.2813,s(h˜4)=0.0802.
Thus, the ranking order of projects is A3≻A1≻A2≻A4.
In this example, when the IVIHFHSWGM operator is adopted. The scores of the projects are
s(h˜1)=0.2110,s(h˜2)=0.0997,s(h˜3)=0.2789,s(h˜4)=0.0848,
and the ranking order of projects is A3≻A1≻A4≻A2, which is different from the above ranking order. However, the same optimal choice is derived.
In this example, when Joshi and Kumar’s method (Joshi and Kumar, 2016) is used, similar to Joshi and Kumar (2016), let WC1=0.4,WC2=0.3,WC3=0.3, and WC4=0.4. From formula (3), we get λ=−0.6359. According to formula (2), we can derive the value of any coalition. Using Joshi and Kumar’s algorithm, we derive the following closeness coefficients of the projects:
Cc1=0.8653,Cc2=0.8843,Cc3=0.8810,Cc4=0.8779,
by which the ranking order is A2≻A3≻A4≻A1, where the ranking order as well as the best choice are different from that obtained above.
These two methods both consider the interactive characteristics between the weights of elements in a set. When we assume that there is no interaction, using models (19) and (22) in the setting of additive measures, the weight vectors on the criteria set and on the ordered set are obtained as follows:
WC=(0.1,0.4,0.25,0.25)andWN=(0.1,0.4,0.4,0.1).
Then, we can use the defined aggregation operators in Zhang (2013) to calculate the comprehensives and to rank projects. To show the ranking values and the ranking orders obtained from different methods, please see Table 3.
Ranking values and ranking orders obtained from different methods.
Methods
Ranking values A1, A2, A3, A4
Ranking orders
Our method using the IVIHFHSWA operator
0.2072
0.1075
0.2813
0.0802
A3≻A1≻A2≻A4
Our method using the IVIHFHSWGM operator
0.2110
0.0997
0.2789
0.0848
A3≻A1≻A4≻A2
Joshi and Kumar’s method (Joshi and Kumar, 2016)
0.8653
0.8843
0.8810
0.8779
A2≻A3≻A4≻A1
Zhang’s method using the IVIHFWA operator (Zhang, 2013)
0.2243
0.2646
0.3562
0.2129
A3≻A2≻A1≻A4
Zhang’s method using the IVIHFWG operator (Zhang, 2013)
0.1754
0.1142
0.2783
0.0932
A3≻A1≻A2≻A4
Zhang’s method using the IVIHFOWA operator (Zhang, 2013)
0.1732
0.2513
0.2583
0.1858
A3≻A2≻A4≻A1
Zhang’s method using the IVIHFOWG operator (Zhang, 2013)
0.1389
0.1265
0.2110
0.0894
A3≻A1≻A2≻A4
Zhang’s method using the IVIHFHA operator (Zhang, 2013)
0.1960
0.2410
0.2431
0.2051
A3≻A2≻A4≻A1
Zhang’s method using the IVIHFHG operator (Zhang, 2013)
-0.1969
-0.2394
-0.2217
-0.2470
A1≻A3≻A2≻A4
Note: the IVIHFWA operator: the interval-valued intuitionistic hesitant fuzzy weighted averaging operator, the IVIHFWG operator: the interval-valued intuitionistic hesitant fuzzy weighted geometric operator, the IVIHFOWA operator: the interval-valued intuitionistic hesitant fuzzy ordered weighted averaging operator, IVIHFOWG operator: the interval-valued intuitionistic hesitant fuzzy ordered weighted geometric operator, the IVIHFHA operator: the interval-valued intuitionistic hesitant fuzzy hybrid averaging operator, the IVIHFHG operator: the interval-valued intuitionistic hesitant fuzzy hybrid geometric operator.
To further show the application of the new method and to compare with the previous ones, we provide the following decision making problem.
Let us consider an investment company that wants to invest a sum of money in the best option (Liu and Jin, 2012). There is a panel with four possible alternatives in which to invest the money: A1 is a car company; A2 is a computer company; A3 is a TV company; A4 is a food company. The investment company must make a decision according to the following four attributes: the risk index C1; the growth index C2; the social-political impact index C3; the environmental impact index C4. The evaluating IVIHFE matrix H˜=(h˜ij)4×4 is offered as shown in Table 4.
Let WC=([0.3,0.4],[0.15,0.25],[0.2,0.25],[0.25,0.3]) be the known weight information on the attribute set, and let WN=([0.2,0.3],[0.25,0.4],[0.25,0.4],[0.2,0.3]) be the known weight information on the ordered set.
Because the criterion c2 is benefit, and the criteria c1, c3, and c4 are cost, we need to normalize the IVIHFE matrix H˜ into the following IVIHFE H˜′, please see Table 5.
Similar to Example 3, with respect to different methods the ranking results are obtained as shown in Table 6.
Ranking values and ranking orders obtained from different methods.
Methods
Ranking values A1, A2, A3, A4
Ranking orders
Our method using the IVIHFHSWA operator
−0.0030
−0.1146
−0.0318
−0.0059
A1≻A4≻A3≻A2
Our method using the IVIHFHSWGM operator
0.2110
0.0997
0.2789
0.0848
A1≻A4≻A3≻A2
Joshi and Kumar’s method (Joshi and Kumar, 2016)
0.8653
0.8843
0.8810
0.8779
A1≻A4≻A3≻A2
Zhang’s method using the IVIHFWA operator (Zhang, 2013)
0.2243
0.2646
0.3562
0.2129
A4≻A1≻A3≻A2
Zhang’s method using the IVIHFWG operator (Zhang, 2013)
0.1754
0.1142
0.2783
0.0932
A1≻A4≻A3≻A2
Zhang’s method using the IVIHFOWA operator (Zhang, 2013)
0.1732
0.2513
0.2583
0.1858
A1≻A4≻A3≻A2
Zhang’s method using the IVIHFOWG operator (Zhang, 2013)
0.1389
0.1265
0.2110
0.0894
A1≻A3≻A4≻A2
Zhang’s method using the IVIHFHA operator (Zhang, 2013)
0.1960
0.2410
0.2431
0.2051
A4≻A1≻A3≻A2
Zhang’s method using the IVIHFHG operator (Zhang, 2013)
-0.1969
-0.2394
-0.2217
-0.2470
A1≻A4≻A3≻A2
Examples 3 and 4 both show that different ranking orders might be derived with respect to the different methods that are based on the different aggregation operators and the different operational laws. Thus, when the decision makers make decisions for some problem, they need to first choose the used method. Because the new operational laws avoid the limitations defined in Section 2.2 and the new aggregation operators globally consider the interactions between the weights of the criteria and the weights of the ordered positions, we suggest the decision makers to apply the new method that can also address the situation where the weight information is not exactly known.
To show the differences between our method and two previous ones intuitively, according to their principles, Table 7 is offered.
The comparison of our method and two previous ones.
Our method
Method in Joshi and Kumar (2016)
Method in Zhang (2013)
Can the used operational laws preserve the monotonicity?
Yes
No
No
Are the weights of criteria for the different alternatives the same?
Yes
No
Yes
Are the interactive characteristics between the weights considered?
Yes
Yes
No
Can the complementary, redundant and independent characteristics between the weights of elements be reflected simultaneously?
Yes
No
No
Is the situation where the weighting information is incompletely known considered?
Yes
No
No
Conclusion
As well known, for a given decision making problem, there are main three procedures that need to be solved: determining the weight information, calculating the alternatives’ comprehensive values, and ranking the alternatives. Considering the previous researches about decision making with interval-valued intuitionistic hesitant fuzzy sets, we wrote this paper and proposed two new aggregation operators that overall consider the interactions between elements in a set. To cope with the situation where the weight information is not exactly known, models for the optimal fuzzy measures and 2-additive measures are constructed. Then, we introduced a new decision-making method and offered a practical decision-making problem to show the concrete application of the new theoretical results. Meanwhile, comparison analysis is performed.
This paper focuses on the theoretical research, and we will continue to study the application of the new approach in some other fields including the application in new product screening, propulsion system selection problem, selecting suitable hotels, economic production problem, evaluating machine tool, social media, green supplier selection, and evaluation of the professor in a university. Furthermore, we shall research decision making with other types of hesitant fuzzy sets and dynamic decision making models in the setting of hesitant fuzzy environment.
Acknowledgements
This work was supported by the National Natural Science Foundation of China (Nos. 71273085, 71571192, 71774177, and 71373074), the National Social Science Foundation of China (No. 16BJY119), the Innovation-Driven Planning Foundation of Central South University (No. 2016CXS027), the State Key Program of National Natural Science of China (No. 71431006), the Projects of Major International Cooperation NSFC (No. 71210003), the Hunan Province Foundation for Distinguished Young Scholars of China (No. 2016JJ1024), and the China Postdoctoral Science Foundation (No. 2016M602170).
ReferencesAtanassov, K. (1983). Intuitionistic fuzzy sets. In: , Sofia.Atanassov, K., Gargov, G. (1989). Interval valued intuitionistic fuzzy sets. , 31, 343–349.Beliakov, G. (2005). Learning weights in the generalized OWA operators. , 4, 119–130.Beliakov, G., Janes, S. (2011). Citation-based journal ranks: the use of fuzzy measures. , 167, 101–119.Beliakov, G., Janes, S. (2013). On extending generalized Bonferroni means to Atanassov orthopairs in decision making contexts. , 211, 84–98.Beliakov, G., Calvo, T., Janes, S. (2014). Consensus measures constructed from aggregation functions and fuzzy implications. , 55, 1–8.Chen, S.M., Huang, Z.C. (2017). Multiattribute decision making based on interval-valued intuitionistic fuzzy values and particle swarm optimization techniques. , 397, 206–218.Chen, N., Xu, Z., Xia, M.M. (2013). Interval-valued hesitant preference relations and their applications to group decision making. , 37, 528–540.Fujimoto, K., Kojadinovic, I., Marichal, J.L. (2006). Axiomatic characterizations of probabilistic and cardinal- probabilistic interaction indices. , 55, 72–99.Garg, H. (2017). Novel intuitionistic fuzzy decision making method based on an improved operation laws and its application. , 60, 164–174.Grabisch, M. (1995). Fuzzy integral in multicriteria decision making. , 69, 279–298.Grabisch, M. (1996). The application of fuzzy integrals in multicriteria decision making. , 89, 445–456.Grabisch, M. (1997). k-order additive discrete fuzzy measures and their representation. , 92, 167–189.Gou, X.J., Xu, Z.S. (2017). Exponential operations for intuitionistic fuzzy numbers and interval numbers in multi-attribute decision making. , 16, 183–204.He, Y.D., He, Z., Shi, L.X., Meng, S.S. (2016). Multiple attribute group decision making based on IVHFPBMs and a new ranking method for interval-valued hesitant fuzzy information. , 99, 63–77.Joshi, D., Kumar, S. (2016). Interval-valued intuitionistic hesitant fuzzy Choquet integral based TOPSIS method for multi-criteria group decision making. , 248, 183–191.Jin, F.F., Ni, Z.W., Chen, H.Y., Li, Y.P., Zhou, L.G. (2016). Multiple attribute group decision making based on interval-valued hesitant fuzzy information measures. , 101, 103–115.Lin, J., Jiang, Y. (2014). Some hybrid weighted averaging operators and their application to decision making. , 16, 18–28.Liu, P.D., Jin, F. (2012). Methods for aggregating intuitionistic uncertain linguistic variables and their application to group decision making. , 205, 58–71.Liu, Y.J., Wu, J., Liang, C.Y. (2017a). Some Einstein aggregating operators for trapezoidal intuitionistic fuzzy MAGDM and application in investment evolution. , 32, 63–74.Liu, W., Dong, Chiclana F, Y., Cabrerizo, F.J., Herrera-Viedma, E. (2017b). Group decision-making based on heterogeneous preference relations with self-confidence. In: . https://doi.org/10.1007/s10700-016-9254-8.Meng, F.Y., Chen, X.H. (2014). An approach to interval-valued hesitant fuzzy multi-attribute decision making with incomplete weight information based on hybrid Shapley operators. , 25, 617–642.Meng, F.Y., Chen, X.H. (2017). Correlation coefficient of interval-valued intuitionistic uncertain linguistic sets and its application. , 48, 114–135.Meng, F.Y., Tang, J. (2013). Interval-valued intuitionistic fuzzy multi-criteria group decision making based on cross entropy and Choquet integral. , 28, 1172–1195.Meng, F.Y., Wang, C., Chen, X.H., Zhang, Q. (2016). Correlation coefficients of interval-valued hesitant fuzzy sets and their application based on Shapley function. , 31, 17–43.Meng, F.Y., Tan, C.Q., Chen, X.H. (2017a). Multiplicative consistency analysis for interval reciprocal preference relations: a comparative study. , 68, 17–38.Meng, F.Y., Lin, J., Tan, C.Q., Zhang, Q. (2017b). A new multiplicative consistency based method for decision making with triangular fuzzy reciprocal preference relations. , 315, 1–25.Meng, F.Y., An, Q.X., Tan, C.Q., Chen, X.H. (2017c). An approach for group decision making with interval fuzzy preference relations based on additive consistency and consensus analysis. , 47, 2069–2082.Meng, F.Y., Tang, J., An, Q.X., Chen, X.H. (2017d). Decision making with intuitionistic linguistic preference relations. . https://doi.org/10.1111/itor.12383.Merigo, J.M., Casanovas, M. (2009). The induced generalized hybrid averaging operator and its application in financial decision making. , 2, 95–101.Pérez, I.J., Cabrerizo, F.J., Herrera-Viedma, E. (2010). A mobile decision support system for dynamic group decision making problems. , 40, 1244–1256.Pérez, I.J., Cabrerizo, F.J., Alonso, S., Herrera-Viedma, E. (2014). A new consensus model for group decision making problems with non homogeneous experts. , 44, 494–549.Perez, L.G., Mata, F., Chiclana, F., Kou, G., Herrera-Viedma, E. (2016). Modelling influence in group decision making. , 20, 1653–1665.Shapley, L.S. (1953). A value for n-person game. In: Kuhn, H., Tucker, A. (Eds.), . Princeton University Press, Princeton.Sugeno, M. (1974). Theory of Fuzzy Integral and Its Application. Doctorial Dissertation, Tokyo Institute of Technology.Tan, C.Q. (2011). A multi-criteria interval-valued intuitionistic fuzzy group decision making with Choquet integral-based TOPSIS. , 38, 3023–3033.Tan, C.Q., Wu, D.D., Ma, B.J. (2011). Group decision making with linguistic preference relations with application to supplier selection. , 38, 14382–14389.Torra, V. (2010). Hesitant fuzzy sets. , 25, 529–539.Ureña, M.R., Chiclana, F., Fujita, H., Herrera-Viedma, E. (2015). Confidence-consistency driven group decision making approach with incomplete reciprocal intuitionistic preference relations. , 89, 86–96.Wang, C.Y., Chen, S.M. (2017). Multiple attribute decision making based on interval-valued intuitionistic fuzzy sets, linear programming methodology, and the extended TOPSIS method. , 397, 155–167.Wu, J., Chiclana, F., Fujita, H., Herrera-Viedma, E. (2017). A visual interaction consensus model for social network group decision making with trust propagation. , 122, 39–50.Xu, Z.S. (2004). Uncertain linguistic aggregation operators based approach to multiple attribute group decision making under uncertain linguistic environment. , 166, 171–184.Xu, Z.S. (2010). Choquet integrals of weighted intuitionistic fuzzy information. , 180, 726–736.Xu, Z.S., Chen, J. (2007). On geometric aggregation over interval-valued intuitionistic fuzzy information. In: , Haikou, China, August 24–27, Vol. 2, pp. 466–471.Xu, Z.S., Da, Q.L. (2002). Combined weighted geometric averaging operator and its application. , 32, 506–509.Zadeh, L.A. (1965). Fuzzy sets. , 8, 338–353.Zhang, Z. (2013). Interval-valued intuitionistic hesitant fuzzy aggregation operators and their application in group decision-making. , 2013, 1–33.