Issuu

SS symmetry Article

Multicriteria Decision Making Based on Generalized Maclaurin Symmetric Means with Multi-Hesitant Fuzzy Linguistic Information Peide Liu 1, * 1 2

and Hui Gao 1,2

School of Management Science and Engineering, Shandong University of Finance and Economics, Jinan 250014, China; gaohui@mail.sdufe.edu.cn School of Business, Heze University, Heze 274015, China Correspondence: peide.liu@gmail.com or liupd@sdufe.edu.cn

Received: 26 February 2018; Accepted: 21 March 2018; Published: 26 March 2018

Abstract: In multicriteria decision making (MCDM), multi-hesitant fuzzy linguistic term sets (MHFLTSs) can eliminate the limitations of hesitant fuzzy linguistic term sets (HFLTSs) and hesitant fuzzy linguistic sets (HFLSs), and emphasize the importance of a repeated linguistic term (LT). Meanwhile, there is usually an interrelation between criteria. The Maclaurin symmetric mean (MSM) operator can capture the interrelationships among multi-input arguments. The purpose of this paper is to integrate MHFLTSs with MSM operators and to solve MCDM problems. Firstly, we develop the generalized MSM operator for MHFLTSs (MHFLGMSM), the generalized geometric MSM operator for MHFLTSs (MHFLGGMSM), the weighted generalized MSM operator for MHFLTSs (WMHFLGMSM) and the weighted generalized geometric MSM operator for MHFLTSs (WMHFLGGMSM), respectively. Then, we discuss their properties and some special cases. Further, we present a novel method to deal with MCDM problems with the MHFLTSs based on the proposed MSM operators. Finally, an illustrative example about how to select the best third-party logistics service provider is supplied to demonstrate the practicality and reliability of the proposed approaches in comparison with some existing approaches. Keywords: multi-hesitant fuzzy linguistic term sets (MHFLTSs); maclaurin symmetric mean (MSM); multicriteria decision making (MCDM)

1. Introduction Because human preferences are inherently vague, it is more suitable to deal with uncertainty and imprecise information by linguistic terms (LTs). For example, when evaluating customer satisfaction of a service or product, experts are inclined to select some LTs such as “poor”, “good” or “excellent”, etc., to give their evaluation of the criteria. Since Zadeh [1] presented the fuzzy linguistic technique in 1975, which utilized linguistic variables (LVs) to describe qualitative information, some methods based on LVs have been utilized in a number of areas, such as supply chain management [2], qualify function deploy [3], health care system [4], housing marker [5], risk evaluation [6], etc. Generally, if the qualitative information is only represented by a single LT, it might sometimes not accurately reflect what it means. When the decision-making problems become more ambiguous, the experts might be hesitant among several LTs and need two or more LTs to express their preference. Based on hesitant fuzzy set (HFS), Rodríguez et al. [7,8] defined a hesitant fuzzy LT set (HFLTS), which is a set with ordered consecutive LTs. In real applications, experts can give several possible LTs instead of a single LT to evaluate the criteria. Although HFLTS improves the previous linguistic approaches, it is still difficult to express the hesitance preference of experts only by consecutive LTs.

Symmetry 2018, 10, 81; doi:10.3390/sym10040081

www.mdpi.com/journal/symmetry

Symmetry 2018, 10, 81

2 of 25

In addition, there are some limitations in the calculation of HFLTS and this will be illustrated in Section 2 of this paper. Thus, hesitant fuzzy linguistic sets (HFLSs), as an extension of HFLTS, were produced, which allow for inconsecutive LTs. However, in practice, when multi-persons give the same LT in assessment information, this LT is handled as one time by default and the significance of repeated LTs is neglected. Since the evaluation values { x1 , x2 , x3 }, { x1 , x2 , x2 , x3 } and { x1 , x2 , x2 , x2 , x3 } are not equivalent to each other, the repeated occurrence of x2 can convey importance or a certain special meaning. In addition, the LVs provided by experts generally fluctuate in evaluation period [9]. So, MHFLTS [10] was proposed based on multi-HFSs to solve the mentioned problems, and each multi-hesitant LT element (MHFLTE) can contain inconsecutive and repeated LTs. Based on the MHFLTS, one expert can assess an alternative under specific criteria by one or several arbitrary LTs, and the assessment values from the different experts, can be collected, and the frequency of a LT can be expressed in the assessment information [11]. In a word, MHFLTS is more energetic than the HFLTS and HFLS, and it can be applied to present the hesitance linguistic information by multi-repeated LTs in primary assessment information. To date, many contributions have concentrated on the decision-making techniques based on HFLTSs, which are from three domains: (1) the theory of foundations, for instance, operational laws [7,8], comparative methods [12,13], distance and similarity measures [14], likelihood [15], outranking degree [16], consistency [17] or multiplicative consistency [18],correlation coefficient [19] and so on; (2) the extended multicriteria decision making (MCDM) approaches for HFLTS, such as TOPSIS [20], ELECTRE [21], VIKOR [22], TODIM [23], Entropy [24], GRA [25] and other methods, such as MACBETH [26], linear programming technique [27],Shapely [28], EDAS [29], Fuzzy Petri Net [30] and so on; (3) the MCDM techniques based on aggregation operators of HFLS. The MCDM methods based on aggregation operators can acquire the comprehensive values of alternatives by aggregating all attribute values, and then rank the alternatives. Obviously, they have more superiority than the traditional MCDM methods, and it is meaningful to research the aggregation operators and then to solve the MCDM problems. As an important tool for information fusion, a great number of research achievements about aggregation operators of HFLS have been produced. Wang et al. [31] presented a 2-tuple linguistic aggregation operator for MHFLTEs; Wu [32] applied possibility distribution to develop the weighted average (WA) operator for HFLSs (HFLWA) and the ordered WA operator (OWA) for HFLSs (HFLOWA). Gou et al. [33] developed the hesitant fuzzy linguistic BM operator for HFLSs (HFLBM) and the weighted HFLBM operator (WHFLBM). Zhu et al. [34] investigated some linguistic hesitant fuzzy power aggregation (LHFPA) operators. Wang et al. [35] proposed the cloud WA (CWA) operator, cloud OWA (COWA) operator, and cloud hybrid arithmetic (CHA) operator for LVs. Liu et al. [36] investigated prioritized WA operator for HIFLSs (HIFLPWA) and prioritized weighted geometric operator for HIFLSs (HIFLPWG). However, these operators for HFLTSs do not pay attention to the interrelation among input arguments. MSM was proposed by Maclaurin [37] and was further developed by Detemple and Robertson [38], which has distinct advantage, i.e., it can deal with the interrelationship among multi-inputs, however Bonferroni mean (BM) operator and Heronian mean (HM) operator can only consider the correlation between two arguments. Consequently, based on its main advantage of more flexible and robust in information fusion, the related research achievements of MSM operator have been quite fruitful. Liu and Zhang [39] introduced some MSM operator for single-valued trapezoidal neutrosophic numbers. Yu et al. [33] developed the MSM operator for HFLSs and the weighted MSM operator for HFLSs. Liu and Qin [40] extended the MSM operator for linguistic intuitionistic fuzzy numbers (LIFNs). Qin and Liu [41] investigated the dual MSM (DMSM) operator and extended the DMSM operator to uncertain LVs. Moreover, the MSM operators were also applied to handle the 2-tuple linguistic information [42] and single-valued neutrosophic linguistic information [43] and so on. However, research about MHFLTSs and their application in MCDM is limited. Wang et al. [31] presented a generalized 2-tuple linguistic WA operator and a generalized 2-tuple linguistic OWA to deal

Symmetry 2018, 10, 81

3 of 25

with MHFLTSs. Afterwards, a likelihood-based TODIM method based on the MHFLTSs for evaluation in logistics outsourcing was also proposed [11]. Then, Wang et al. [10] further developed the HM and prioritized operators to solve the MCDM problem with the MHFLTSs. Liu and Teng [44] proposed some normal neutrosophic number Heronian Mean operators for solving multiple attribute group decision making problems. However, because MHFLTSs can describe inconsecutive and repeated LTs and MSM aggregation operator can process the correlations among multi-inputs, to develop some MSM operators to deal with MHFLTSs is an important work. So, the aim of this paper is to establish some MSM operators for MHFLTSs and use them to solve the MCDM problem in which the attributes take the form of MHFLTSs. This paper is organized as follows. Section 2 reviews and discusses some basic concepts and theories. Section 3 proposes some multi-hesitant fuzzy linguistic generalized MSM operators, including the generalized MSM operator for MHFLTSs (MHFLGMSM), the generalized geometric MSM operator for MHFLTSs (MHFLGGMSM), the weighted generalized MSM operator for MHFLTSs (WMHFLGMSM) and the weighted generalized geometric MSM operator for MHFLTSs (WMHFLGGMSM), and studies some properties and some particular examples of these operators. Section 4, outlines an MCDM approach based on the proposed aggregation operators. Section 5 gives a case to verify the availability of the presented methods. Section 6 presents a few conclusions. 2. Preliminaries In this part, the definitions of HFLTSs, HFLSs and MHFLTEs are briefly reviewed and their corresponding operations are given. Subsequently, the linguistic scale function (LSF) and score function of MHFLTS were given. 2.1. HFLTSs Assume that S = { si |i = 0, 1, 2, · · · , 2g, g ∈ N } is a LT set (LTs) with odd cardinality, where si is called a LV. Then the following requirements must be satisfied [2]: (1)

The set is ordered: α > β ⇔ sα > s β ;

(2)

There is a negation operator: neg(sα ) = s2g−α .

Definition 1. Let S = { si |i = 0, 1, 2, · · · , 2g, g ∈ N }, then a HFLTS Hs is an ordered finite consecutive LTs of S [31]. Definition 2. Let Hs , Hs1 and Hs2 be any three HFLTSs on S [31]. Then we have (1) (2) (3)

The upper bound hs+ : hs+ = max{ hτ |hτ ∈ Hs }; and the lower bound hs− of Hs : hs− = min{ hτ |hτ ∈ Hs } and hs− ≤ hs+ ; The intersection between Hs1 and Hs2 : Hs1 ∩s Hs2 = hτ |hτ ∈ Hs1 and hτ ∈ Hs2 ; The union of Hs1 and Hs2 : Hs1 ∪s Hs2 = hτ |hτ ∈ Hs1 or hτ ∈ Hs2 .

Example 1. Assume that S = { si |i = 0, 1, 2, · · · , 8}, Hs1 = {s5 , s6 }, Hs2 = {s3 , s4 , s5 } and Hs3 = {s1 , s2 , s3 } are three HFLTSs on S. Then, we get: (1) (2)

h1s+ = {s6 } and h1s− = {s5 }; Hs1 ∩s Hs2 = {s5 } and Hs1 ∪s Hs3 = {s1 , s2 , s3 , s5 , s6 }.

Obviously, the union of Hs1 and Hs3 is not a HFLTS because the HFLTS must be consecutive LTs. So, the HFLTSs were extended to HFLSs.

Symmetry 2018, 10, 81

4 of 25

2.2. HFLSs In order to preserve the provided information, the LTs S is developed to a successive LTs S = { sk |k ∈ [0, 2g]}. If sk ∈ S, then sk is named the original LT; if not so, sk is named the virtual LT which has no practical meaning and only it is used to operational process. Definition 3. Let X be a reference set and S = { sk |k ∈ [0, 2g]}. Then a HFLS HLS on X can be defined in terms of a function HLS that returns a subset of S [32]. For simple and convenient, h LS ( x ) = h LS is called a HFLE. Definition 4. Let S = { sk |k ∈ [0, 2g]} be a LTs, h LS , h1LS and h2LS be three HFLEs on S and λ ∈ [0, 1], I (sk ) = k [32]. Thus, the operations can be defined as follows: n

h1LS ⊕ h2LS = ∪h

∈h1LS ,hτ2 ∈h2LS τ1

{ hτ 1 ⊕ hτ2 } = ∪h

∈h1LS ,hτ2 ∈h2LS τ1

h1LS ⊗ h2LS = ∪h

∈h1LS ,hτ2 ∈h2LS τ1

{ hτ 1 ⊗ hτ2 } = ∪h

∈h1LS ,hτ2 ∈h2LS τ1

n n

s I (h

τ1

s I (h

)× I (hτ2 ) τ1

λh LS = ∪hτ ∈h LS {λhτ } = ∪hτ ∈h LS s(λI (hτ )) n o n o (h LS )λ = ∪hτ ∈h LS (hτ )λ = ∪hτ ∈h LS s( I (h ))λ τ

(1)

(2)

)+ I (hτ2 )

(3) (4)

Example 2. Suppose that S = {sk |k ∈ [0, 8]} and h1LS , h2LS , h3LS are three HFLEs on S. h1LS = {s5 , s6 }, h2LS = {s3 , s4 , s5 } and h3LS = {s1 , s2 , s3 }. Then, h1LS ⊕ h2LS = {s8 , s9 , s10 , s11 }, h1LS ⊗ h3LS = {s5 , s6 , s10 , s12 , s15 , s18 }. The results obtained by Definition 4 have exceed the domain of S. Many elements based on the subscript of LTs in h1LS ⊕ h2LS and h1LS ⊗ h3LS do not exist in S.

Definition 5. Let S = {sk |k ∈ [0, 2g]} be a LTs and h LS = hτ hτ ∈ S be a HFLE on S [33]. The score function of h LS is 1 I ( hτ ) S(h LS ) = (5) #h LS h ∑ 2g ∈h τ

where #h LS indicates the count of elements in h LS . h1LS , h2LS h1LS > h2LS , and if S h1LS = S h2LS then h1LS = h2LS .

are two arbitrary HFLEs, if S h1LS > S h2LS then

Example 3. Assume that S = {sk |k ∈ [0, 8]} and h1LS , h2LS are two HFLEs on S. h1LS = {s2 , s4 , s6 }, h2LS = {s2 , s6 }, S h1LS = S h2LS = 0.5, then h1LS = h2LS . However, this comparison result is insufficient due to the existence of s4 . Therefore, it is essential to give some revisions. 2.3. MHFLTS Although the definitions of HFLTS and HFLS have been widely and consistently used, the frequency of one repeated LT is assigned to be one time and the operations are contradictory to the definitions. Thus, a further extended definition of them is proposed in this section. Definition 6. Suppose X is a reference set, a multi-HFS on X can be defined about a function h E that returns a multi-subset of values in [0, 1] [31]. Definition 7. Suppose X is a reference set and S = {sk |k ∈ [0, 2g]} is a LTs. Then a MHFLTS H on X is defined about a function h that returns an ordered finite multi-subset of Ŝ, denoted by [31]:

Symmetry 2018, 10, 81

5 of 25

H = {h x, h( x )i| x ∈ X }

(6)

where h(x) indicates the possible membership degrees of the element x ∈ X to the set X. For simple and convenient, h(x) = h is a MHFLTE and Ĥ is the set of all MHFLTEs. There is the strong possibility that the virtual LTs of h exist in S but not in S. Essentially, MHFLTSs are a development of HFLTSs and HFLSs. If h is a set of successive LTs and no duplicate LT, then h is a HFLTSs. If h is a set of discrete LTs with no duplicate LT, then h is a HFLS. Therefore, all the operations and approaches on MHFLTSs can be applied to HFLTSs and HFLSs because they are both special cases of MHFLTSs. Definition 8. Let S = {sk |k ∈ [0, 2g]} be a LTs. The LSF ϑ : S → R+ is a strictly monotone increasing and continuous function satisfying ϑ (0) = 0 and ϑ (2g) = 1. The inverse function of ϑ exists and is denoted by ϑ−1 (·). Then ϑ−1 (δ) = hτ , where hτ ∈ S and δ ∈ R+ [41,42]. There are five normal LSFs which are shown as below. ϑ ( hτ ) = ϑ ( hτ ) = ϑ ( hτ ) =

I ( hτ ) 2g

I (hτ ) ∈ [0, 2g]

I ( hτ ) 2g

I (hτ ) ∈ [0, 2g]

(8)

I (hτ ) ∈ [0, 2g]

(9)

 g a − a g− I (hτ )    2( a g − 1) ϑ ( hτ ) = g I (h )− g − 2   a +a τ  2( a g − 1)  a g − ( g − I (hτ )) a    2gα ϑ ( hτ ) = β  g + ( I ( hτ ) − g) β   2gα

(7)

( I (hτ ) ∈ [0, g]) (10)

( I (hτ ) ∈ [0, g])

( I (hτ ) ∈ [0, g]) ,

where α, β ∈ (0, 1]

(11)

( I (hτ ) ∈ [0, g])

A great quantity of experiment research has proven the value of a in Equation (10) lies in the interval [1.36, 1.4]. Suppose g = 4, a = 1.4 and α = β = 0.8, the character of Equations (7)–(11) can be shown in Figure 1.

Figure 1. The graphical demonstration of Equations (7)–(11).

Symmetry 2018, 10, 81

6 of 25

The Equation (10) is regarded as one representative among all Equations for ϑ−1 (·), thus we choose it to deal with the MHFLTEs in this paper. 2.4. Basic Operations of MHFLTS Definition 9. Let h1 and h2 be any two MHFLTEs on S. Then, we can give the following operations.

o n

h1 ⊕ h2 = ϑ−1 (ϑ (hτ1 ) + ϑ (hτ2 )) hτ1 ∈ h1 , hτ2 ∈ h2

o n

h1 ⊗ h2 = ϑ−1 (ϑ (hτ1 ) + ϑ (hτ2 )) hτ1 ∈ h1 , hτ2 ∈ h2

o n

λh1 = ϑ−1 (λϑ (hτ )) hτ ∈ h1

o n

(h1 )λ = ϑ−1 (ϑ(hτ ))λ hτ ∈ h1

(12) (13) (14) (15)

Definition 10. Let h, h1 , h2 be arbitrary MHFLTEs on S. Then we give the score function of h as follows [31]: 1 ϑ ( hτ ) #h h∑ ∈h

(16)

1 2 (ϑ(hτ ) − S(h)) #h h∑ ∈h

(17)

S(h) =

and the variance function of h as follows: V (h) =

where #h denotes the count of elements in h. Then the comparison rules for MHFLTEs can be re-defined as follows: (1) If ϑ h1− > ϑ h1+ , then h1 is strictly greater than h2 , denoted by h1 > h2 ; (2) If S(h1 ) > S(h2 ), then h1 is greater than h2 , denoted by h1 > h2 ; If S(h1 ) = S(h2 ), and V (h1 ) < V (h2 ), then h1 > h2 ; If S(h1 ) = S(h2 ), and V (h1 ) = V (h2 ), then h1 = h2 . 2.5. MSM Operator Definition 11. Let ai (i = 1, 2, · · · , n) be a group of positive real values and m = 1, 2, · · · , n [43]. If

MSM

(m)

( a1 , a2 , · · · , a n ) =

∑1≤i1 <···<im ≤in ∏m j =1 a i j

Cnm

(18)

Then MSM(m) is called the MSM operator where (i1 , i2 , · · · , im ) traversal all the combination of (1, 2, · · · , n), Cnm is the binominal coefficient. It is clear that the MSM(m) has the following properties: (1)

Idempotency. MSM(m) ( a, a, · · · , a) = a;

(2)

Monotonicity. MSM(m) ( a1 , a2 , · · · , an ) ≤ MSM(m) (b1 , b2 , · · · , bn ), if ai < bi for all i;

(3)

Boundedness. min{ a1 , a2 , · · · , an } ≤ MSM(m) ( a1 , a2 , · · · , an ) ≤ max{ a1 , a2 , · · · , an }. Especially, if m = 1, then the MSM(1) reduces to the WA operator as follows: MSM(1) ( a1 , a2 , · · · , an ) =

∑ 1≤ i1 ≤ i n a i n

(19)

Symmetry 2018, 10, 81

7 of 25

If m = 2, then the MSM(2) reduces to a BM operator (p = q = 1), as follows: MSM(2) ( a

1 , a2 , · · ·

n ∑i,j =1,i 6= j ai1 ai2

, an ) =

1 2

n ( n −1)

∑1≤i1 <i2 ≤in ∏2j=1 ai j

Cn2

BM1,1 ( a

1 , a2 , · · ·

2∑

1≤ i1 < i2 ≤ i n

n ( n −1)

a i1 a i2

12 (20)

, an )

If m = 3, then the MSM(3) reduces to a generalized BM operator (p = q = u = 1) as follows:

MSM(3) ( a

∑1≤i1 <i2 <i3 ≤in ∏3j=1 ai j Cn3

= 1 , a2 , · · · , a n ) = n 1 ∑i,j,r=1,i6= j6=r ai1 ai2 ai3 3 = = GBM1,1,1 ( a1 , a2 , · · · , an ) n(n−1)(n−2)

6∑

a i1 a i2 a i3 n(n−1)(n−2)

1≤ i1 < i2 ≤ i n

13 (21)

Definition 12. Let ai (i = 1, 2, · · · , n) be a set of positive real values and m = 1, 2, · · · , n [43]. If  GMSM(m,p1 ,p2 ,··· ,pm ) ( a1 , a2 , · · · , an ) = 

∑1≤i1 <···<im ≤in ∏m j =1 a i j

p

Cnm



1 1 + p2 +···+ pm

(22)

Then GMSM(m,p1 ,p2 ,··· ,pm ) is called the generalized MSM (GMSM) operator, where p1 , p2 , · · · , pm ≥ 0, (i1 , i2 , · · · , im ) traversal all the combination of (1, 2, · · · , n), Cnm is the binominal coefficient. The GMSM(m,p1 ,p2 ,··· ,pm ) has the following properties: Property 1. (1)

Idempotency. GMSM(m,p1 ,p2 ,··· ,pm ) ( a, a, · · · , a) = a;

(2)

Monotonicity. GMSM(m,p1 ,p2 ,··· ,pm ) ( a1 , a2 , · · · , an ) ≤ GMSM(m,p1 ,p2 ,··· ,pm ) (b1 , b2 , · · · , bn ), if ai ≤ bi for all i; Boundedness. min{ a1 , a2 , · · · , an } ≤ GMSM(m,p1 ,p2 ,··· ,pm ) ( a1 , a2 , · · · , an ) ≤ max{ a1 , a2 , · · · , an }.

(3)

Proof. (1)

GMSM(m,p1 ,p2 ,··· ,pm ) ( a, a, · · · , a) = 1 m p + p +···+ pm p1 + p2 +···+ pm 2 1 = Cn a Cm =a

∑1≤i1 <···<im ≤in ∏m j =1 a Cnm

1 1 + p2 +···+ pm

∑1≤i1 <···<im ≤in a p1 + p2 +···+ pm Cnm

1 p1 + p2 +···+ pm

(2)

Assume that m-tuple (i1 , i2 , · · · , im ) is given randomly, and p1 , p2 , · · · , pm ≥ 0. if 0 ≤ ai ≤ bi for pj

m m m all i, then ∏m j=1 ai j ≤ ∏ j=1 bi j and ∑1≤i1 <···<im ≤in ∏ j=1 ai j ≤ ∑1≤i1 <···<im ≤in ∏ j=1 bi j , therefore,

(3)



∑1≤i1 <···<im ≤in ∏m j =1 a i j

p



Cnm



1 1 + p2 +···+ pm



≤

∑1≤i1 <···<im ≤in ∏m j = 1 bi j

p

Cnm



1 1 + p2 +···+ pm

Let a− = min{ a1 , a2 , · · · , an } and a+ = max{ a1 , a2 , · · · , an }. According to the − ( m,p ,p , ··· ,p m ) ( a − , a − , · · · , a − ). 2 1 property of idempotency, min{ a1 , a2 , · · · , an } = a = GMSM − According to the property of monotonicity, when a ≤ ai for all i, we have ( m,p ,p , ··· ,p ) − − − ( m,p ,p , ··· ,p ) m m 2 2 1 1 GMSM ( a , a , · · · , a ) ≤ GMSM ( a1 , a2 , · · · , an ). Similarly, we also have GMSM(m,p1 ,p2 ,··· ,pm ) ( a1 , a2 , · · · , an ) ≤ GMSM(m,p1 ,p2 ,··· ,pm ) ( a+ , a+ , · · · , a+ ). Finally, MSM(m,p1 ,p2 ,··· ,pm ) ( a1 , a2 , · · · , an ) ≤ MSM(m,p1 ,p2 ,··· ,pm ) (b1 , b2 , · · · , bn ).

Symmetry 2018, 10, 81

8 of 25

Especially, if m = 1, then the GMSM(m,p1 ,p2 ) ( a1 , a2 , · · · , an ) reduces to the generalized WA operator. 1 p!p a ∑ 1 ≤ i ≤ i n i 1 GMSM(1,p) ( a1 , a2 , · · · , an ) = (23) n If m = 2, then the GMSM(m,p1 ,p2 ) ( a1 , a2 , · · · , an ) reduces to the BM operator as follows: GMSM(2,p1 ,p2 ) ( a

p1 p2 n ∑i,j =1,i 6= j ai a j

1 , a2 , · · ·

1 p1 + p2

n ( n −1)

, an ) =

∑1≤ i1 < i2 ≤ i n a i 1 a i 2 Cn2

BM p1 ,p2 ( a

1 , a2 , · · ·

1 1 + p2

2∑1≤ i < j ≤ n a i 1 a j 2

1 1 + p2

n ( n −1)

(24)

, an )

If m = 3, then the GMSM(m,p1 ,p2 ,p3 ) ( a1 , a2 , · · · , an ) reduces to the GBM operator as follows:  GMSM(3,p1 ,p2 ,p3 ) ( a1 , a2 , · · · , an ) = 

p1 p2 p3 n ∑i,j,r =1,i 6= j6=r ai a j ar

n(n−1)(n−2)

1 1 + p2 + p3

∑1≤i1 <i2 <i3 ≤in ∏3j=1 ai j



Cn3



1 p1 + p2 + p3

6∑1≤ i < i ≤ i n a i 1 a i 2 a i 3 1 2 1 2 3 n(n−1)(n−2)

1 1 + p2

(25)

= GBM p1 ,p2 ,p3 ( a1 , a2 , · · · , an )

Definition 13. Let ai (i = 1, 2, · · · , n) be a group of positive real values and m = 1, 2, · · · , n [43]. If

GGMSM

(m,p1 ,p2 ,··· ,pm )

1 ( a1 , a2 , · · · , a n ) = p1 + p2 + · · · + p m

∏

∑ p j ai j

1 m Cn

(26)

1≤i1 <···<im ≤in j=1

Then GGMSM(m,p1 ,p2 ,··· ,pm ) is called the generalized geometric MSM (GGMSM), where p1 , p2 , · · · , pm ≥ 0, (i1 , i2 , · · · , im ) traversal all the combination of (1, 2, · · · , n), Cnm is the binominal coefficient. The GGMSM(m,p1 ,p2 ,··· ,pm ) has the following properties: Property 2. (1)

Idempotency. GGMSM(m,p1 ,p2 ,··· ,pm ) ( a, a, · · · , a) = a;

(2)

Monotonicity. GGMSM(m,p1 ,p2 ,··· ,pm ) ( a1 , a2 , · · · , an ) ≤ GGMSM(m,p1 ,p2 ,··· ,pm ) (b1 , b2 , · · · , bn ), if 0 ≤ ai ≤ bi for all i; Boundedness. min{ a1 , a2 , · · · , an } ≤ GGMSM(m,p1 ,p2 ,··· ,pm ) ( a1 , a2 , · · · , an ) ≤ max{ a1 , a2 , · · · , an }.

(3)

Property 2 is similar to the Property 1, the proof is omitted here. Especially, if m = 1, then the GGMSM(m,p1 ,p2 ,··· ,pm ) reduces to the generalized geometric operator

GGMSM

(1,p)

1 ( a1 , a2 , · · · , a n ) = p

∏ pai

(27)

i =1

If m = 2, then the GGMSM(m,p1 ,p2 ,··· ,pm ) reduces to the geometric BM operator as follows: GGMSM(2,p1 ,p2 ) ( a1 , a2 , · · · , an ) = p +1 p2 ∏1≤i1 <i2 ≤in p1 ai1 + p2 ai2 1 C12 n 1 n = GBM p1 ,p2 ( a , a , · · · , a ) = p1 + p2 ∏i,j=1,i6= j p1 ai + p2 a j n 1 2

1 2 Cn

(28)

Symmetry 2018, 10, 81

9 of 25

If m = 3, then the GGMSM(3,p1 ,p2 ,p3 ) ( a1 , a2 , · · · , an ) reduces to the generalized geometric BM operator as follows: GGMSM(3,p1 ,p2 ,p3 ) ( a1 , a2 , · · · , an ) = p + p12 + p3 ∏1≤i1 <i2 <i3 ≤in p1 ai1 + p2 ai2 + p3 ai3 1 13 n 1 = p + p2 + p3 ∏i,j,r=1,i6= j6=r p1 ai + p2 a j + p3 ar Cn = GGBM p1 ,p2 ,p3 ( a1 , a2 , · · · , an )

1 3 Cn

(29)

3. Some Multi-Hesitant Fuzzy Linguistic MSM Operators In this section, we will extend GMSM operator and GGMSM operator to process MHFLTSs and develop the MHFLGMSM operator, MHFLGGMSM operator, WMHFLGMSM operator and WMHFLGGMSM operator. 3.1. MHFLGMSM Operator Definition 14. Let hi (i = 1, 2, · · · , n) be a group of MHFLTESs on S. A MHFLGMSM operator is a mapping Ĥ n → Ĥ defined as follows:   MHFLGMSM(m,p1 ,p2 ,··· ,pm ) (h1 , h2 , · · · , hn ) = 

⊕1≤τi

<···<τim <n 1

j ⊗m j=1 hτi

p

Cnm

1 1 + p2 +···+ pm

 

(30)

where m = 1, 2, · · · n and Ĥ is the set of all MHFLTESs. Based on the operations of the MHFLTESs described in Section 2, we can get the following theorem. Theorem 1. Let hi (i = 1, 2, · · · , n) be a set of MHFLTESs on S, the value aggregated by the MHFLGMSM operator from (30) is still a MHFLTSs and  1 p j  p + p +···+  m pm 2 1 Cn m    −1  ∑k=1 ∏ j=1 ϑ hτi kj      MHFLGMSM(m,p1 ,p2 ,··· ,pm ) (h1 , h2 , · · · , hn ) = ∪hτ ∈h1 ,hτ ∈h2 ,··· ,hτn ∈hn  ϑ     2 1 Cnm   

(31)

where m = 1, 2, · · · , n and k = 1, 2, . . . , Cnm . hτi k represents the i j th element in kth union of each permutation j

which consists of one element from each hi (i = 1, 2, · · · , n). Because the form of MHFLGMSM(m,p1 ,p2 ,··· ,pm ) involves selecting an element from each hi (i = 1, 2, · · · , n), each hi must mutual calculation and obviously the MHFLGMSM operator will be used ∏(#hi , i = 1, 2, · · · , n) times, where #hi denotes the number of elements in hi . The final aggregated result consists of ∏(#hi , i = 1, 2, · · · , n) elements because of each aggregated result becomes an element which based on the operational laws of MHFLTSs. Property 3. Let hi (i = 1, 2, · · · , n) be a set of MHFLTESs on S. ( m,p 1 ,p2 ,··· ,pm ) operator has the following desirable properties: MHFLGMSM (1)

The value aggregated by

Idempotency. If the hi = h, then S MHFLGMSM(m,p1 ,p2 ,··· ,pn ) (h, h, · · · , h) = S(h)

(2)

Monotonicity. If (h1 , h2 , · · · , hn ) and h10 , h20 , · · · , h0n are two sets of MHFLTESs on S and hi+ < hi0− , then for i = 1, 2, · · · , n,

Symmetry 2018, 10, 81

10 of 25

MHFLGMSM(m,p1 ,p2 ,··· ,pn ) (h1 , h2 , · · · , hn ) < MHFLGMSM(m,p1 ,p2 ,··· ,pn ) h10 , h20 , · · · , h0n (3)

Boundedness. If hmin is obtained by replacing the minimum of hi for each element of hi , hmax is obtained i i by replacing the maximum of hi for each element of hi , then min min ≤ MHFLGMSM (m,p1 ,p2 ,··· ,pn ) ( h , h , · · · , h ) MHFLGMSM(m,p1 ,p2 ,··· ,pn ) hmin n 1 2 1 , h2 , · · · , h n max , · · · , hmax ≤ MHFLGMSM(m,p1 ,p2 ,··· ,pn ) hmax , h n 2 1

(4)

Commutativity. Let h10 , h20 , · · · , h0n be any permutation of (h1 , h2 , · · · , hn ), then MHFLGMSM(m,p1 ,p2 ,··· ,pn ) h10 , h20 , · · · , h0n = MHFLGMSM(m,p1 ,p2 ,··· ,pn ) (h1 , h2 , · · · , hn )

Proof (1)

Since the MHFLTSs hi = h, then  S

MHFLGMSM(m,p1 ,p2 ,··· ,pn ) (h, h, · · ·

= S ∪ hτ

(2)

, h ) = S  ∪ hτ

ϑ−1 (ϑ (hτi ))

∈h1 ,hτ2 ∈h2 ,··· ,hτn ∈hn



∈h1 ,hτ2 ∈h2 ,··· ,hτn ∈hn

(#h)n 1 (#h)n ∑k=1

 ϑ −1

1 ! pj Cm p1 + p2 +···+ pm ∑k=n 1 ∏m j=1 ( ϑ ( hτi )) m Cn

 

ϑ (hτi ) = S(h).

If (h1 , h2 , · · · , hn ) and h10 , h20 , · · · , h0n are two sets of MHFLTESs on S and hi+ < hi0− , then for i = 1, 2, · · · , n S MHFLGMSM(m,p1 ,p2 ,··· ,pn ) (h1 , h2 , · · · , hn )    1 ! p Cm p1 + p2 +···+ pm ϑ ( hτi )) j ∑k=n 1 ∏m ( j = 1  = S∪hτ ∈h1 ,hτ ∈h2 ,··· ,hτn ∈hn ϑ−1 Cm 1



1 #h1 ×#h2 ×···×#hn

∑

hτ1 ∈h1 ,hτ2 ∈h2 ,··· ,hτn ∈hn

∑

hτ1 ∈h1 ,hτ2 ∈h2 ,··· ,hτn ∈hn

ϑ  ϑ −1

1 ! pj Cm p1 + p2 +···+ pm ∑k=n 1 ∏m j=1 ( ϑ ( hτi )) Cnm

 

1 ! pj Cm p1 + p2 +···+ pm ∑k=n 1 ∏m j=1 ( ϑ ( hτi )) Cnm

The hi+ < hi0− , and for i = 1, 2, · · · , n states the minimum element of hi0− is more than the maximum element of hi+ , then ϑ hi+ < ϑ hi0− . In addition, MHFLGMSM operator can satisfy the property of monotonicity, so 1 1  m p j  p + p +···+ p j  p + p +···+ pm pm 1 2 1 2 Cn Cnm m m 0 ϑ h ϑ h ∑ ∑ ∏ ∏ τi τi j =1 j =1  k =1   k =1  <    Cnm Cnm



Finally, MHFLGMSM(m,p1 ,p2 ,··· ,pn ) (h1 , h2 , · · · , hn ) < MHFLGMSM(m,p1 ,p2 ,··· ,pn ) h10 , h20 , · · · , h0n (3)

If hmin is obtained by replacing the minimum of hi for each element of hi and hmax is obtained i i min max by replacing the maximum of hi for each element of hi , then hτi ≤ hτi ≤ hτi and ϑ hmin ≤ τi ϑ (hτi ) ≤ ϑ hmax for hmin ∈ hmin , hτi ∈ hi , hmax ∈ hmax . In addition, the MHFLGMSM operator τi τi τi i i can satisfy the property of monotonicity, so

Symmetry 2018, 10, 81

11 of 25

1 p ! Cm j p1 + p2 +···+ pm min ∑k=n 1 ∏m j=1 ϑ hτi Cnm

1 p ! Cm j p1 + p2 +···+ pm ∑k=n 1 ∏m j=1 ϑ hτi Cnm

≤

1 p ! Cm j p1 + p2 +···+ pm max ∑k=n 1 ∏m j=1 ϑ hτi Cnm

≤

Then 



−1 S∪hmin min ,hmin ∈ hmin ,··· ,hmin ∈ hmin  ϑ n τ τ τ ∈h 1



≤ S  ∪ hτ

∈h1 ,hτ2 ∈h2 ,··· ,hτn ∈hn

∏m j =1

 ϑ −1

1 p ! j p1 + p2 +···+ pm ϑ hmin τ



Cnm



1 ! pj Cm p1 + p2 +···+ pm ∑k=n 1 ∏m j=1 ( ϑ ( hτi )) Cnm



−1 max max max max max  ϑ ≤ S∪hmax τ ∈ h1 ,hτ ∈ h2 ,··· ,hτn ∈ hn 2

Cm ∑k=n 1

 

1 p ! Cm j p1 + p2 +···+ pm max ∑k=n 1 ∏m j=1 ϑ hτi Cnm

 

Finally, min min ≤ MHFLGMSM (m,p1 ,p2 ,··· ,pn ) ( h , h , · · · , h ) MHFLGMSM(m,p1 ,p2 ,··· ,pn ) hmin n 1 2 1 , h2 , · · · , h n max , · · · , hmax . , h ≤ MHFLGMSM(m,p1 ,p2 ,··· ,pn ) hmax n 2 1 (4)

Because h10 , h20 , · · · , h0n is a permutation of (h1 , h2 , · · · , hn ), so ∏(#hi , i = 1, 2, · · · , n) = ∏ #hi0 , i = 1, 2, · · · , n . Therefore pj ⊕1≤τi <···<τim <n ⊗m hτi j = 1 1 j 

 

p

Cnm

1 1 + p2 +···+ pm

 



⊗m j =1

 ⊕1≤τi1 <···<τim <n =  Cnm

h0τi j

pj  p

1 1 + p2 +···+ pm

  

where ∃hτi ∈ hi for h0τi ∈ hi0 . So MHFLGMSM(m,p1 ,p2 ,··· ,pm ) (h1 , h2 , · · · , hn )   Cm ∑k=n 1 ∏m  j =1 ϑ  −1   = ∪hτ ∈h1 ,hτ ∈h2 ,··· ,hτn ∈hn ϑ  Cnm 2 1  

= ∪h0τ

 

∈h10 ,h0τ ∈h20 ,··· ,h0τn ∈h0n  ϑ  2 1

 −1 

 

Cm ∑k=n 1

∏m j =1

ϑ Cnm

!! p !  j

hτ k ij

h0 k τi j

1 p1 + p2 +···+ pm

   

   !! p !  j

  



1 p1 + p2 +···+ pm

    

= MHFLGMSM(m,p1 ,p2 ,··· ,pm ) (h10 , h20 , · · · , h0n ). The proof of Theorem 2 is completed now. Next, by taking into account diverse values of the parameter m, some particular cases of the MHFLGMSM(m,p1 ,p2 ,··· ,pm ) can be discussed as follows:

Symmetry 2018, 10, 81

(1)

12 of 25

When m = 1, then

MHFLGMSM(1,p1 ) (h1 , h2 , · · · , hn ) =

(2)

p ⊕in=1 hi 1

= ∪ hτ



 

∈h1 ,hτ2 ∈h2 ,··· ,hτn ∈hn  ϑ

Cm ∑ n −1  k =1



p1  p1  1 ϑ hτ k   i    n

(32)

When m = 2, then MHFLGMSM(2,p1 ,p2 ) (h

1 , h2 , · · ·



   = ∪hτ ∈h1 ,hτ ∈h2 ,··· ,hτn ∈hn ϑ−1   2 1 

(3)



1 p1

, hn ) =

⊕in=1,j=1,i6= j hi 1 h j 2 n ( n −1)

1 1 + p2

!! p p 2 1 n ϑ h k ∑i=1,j=1,i6= j ϑ h k τ i

τ j

n ( n −1)

   

1 p1 + p2

 (33)    

When m = 3, then 1 n p p p ⊕i=1,j=1,r=1,i6= j6=r hi 1 h j 2 hr 3 p1 + p2 + p3 MHFLGMSM(3,p1 ,p2 ,p3 ) (h1 , h2 , · · · , hn ) = n(n−1)(n−2)   p p  p  p + p1 + p 2 1 3 1 2 3 n ϑ hτ k ϑ hτ k   −1 ∑i=1,j=1,r=1,i6= j6=r ϑ hτi k r j   = ∪hτ ∈h1 ,hτ ∈h2 ,··· ,hτn ∈hn  ϑ    n n − 1 n − 2 ( )( ) 2 1

3.2. MHFLGGMSM Operator Definition 15. Suppose hi (i = 1, 2, · · · , n) is a set of MHFLTESs on S. A MHFLGGMSM operator is a mapping Ĥ n → Ĥ defined as follows: MHFLGGMSM(m,p1 ,p2 ,··· ,pm ) (h1 , h2 , · · · , hn ) =

1m 1 Cn ⊗1≤τi <···<τim <n ⊕m j =1 p j h i j 1 p1 + p2 + · · · + p m

(35)

where m = 1, 2, · · · , n and Ĥ is the set including all MHFLTSs. Based on the operations of MHFLTSs described in Section 2, we can get the following theorem. Theorem 2. Suppose hi (i = 1, 2, · · · , n) is a set of MHFLTESs on S, the value aggregated by MHFLGGMSM operator from (35) is still a MHFLTSs and MHFLGGMSM(m,p1 ,p2 ,··· ,pm ) (h1 , h2 , · · · , hn ) 1m !! m Cn C m 1 n = ∪hτ ∈h1 ,hτ ∈h2 ,··· ,hτn ∈hn ϑ−1 p + p2 +···+ pm ∏k =1 ∑ j=1 p j ϑ hτi k 1

(36)

where m = 1, 2, · · · , n and k = 1, 2, . . . , Cnm . hτi k represents the ij th element in kth union of each permutation j

which consists of one element from every hi (i = 1, 2, · · · , n). Because the form of MHFLGGMSM operator involves selecting an element from every hi (i = 1, 2, · · · , n), each hi must mutual calculation, and obviously the MHFLGGMSM operator will be used ∏(#hi , i = 1, 2, · · · , n) times, where #hi indicates the count of elements in hi . The final aggregated result consists of ∏(#hi , i = 1, 2, · · · , n) elements because of each aggregated result becomes an element which is based on the operation law of MHFLTSs. Property 4. Suppose hi (i = 1, 2, · · · , n) is a group of the MHFLTESs on S. The value aggregated by MHFLGGMSM operator has the following desirable properties:

Symmetry 2018, 10, 81

(1)

13 of 25

Idempotency. If the hi = h, then S MHFLGGMSM(m,p1 ,p2 ,··· ,pn ) (h, h, · · · , h) = S(h)

(2)

Monotonicity. If (h1 , h2 , · · · , hn ) and h10 , h20 , · · · , h0n are two collections of MHFLTSs on S and hi+ < hi0− , then for i = 1, 2, · · · , n. MHFLGGMSM(m,p1 ,p2 ,··· ,pn ) (h1 , h2 , · · · , hn ) < MHFLGGMSM(m,p1 ,p2 ,··· ,pn ) h10 , h20 , · · · , h0n .

(3)

Boundedness. If hmin is obtained by replacing the minimum of hi for each element of hi , hmax is obtained i i by replacing the maximum of hi for each element of hi , then min min ≤ MHFLGGMSM (m,p1 ,p2 ,··· ,pn ) ( h , h , · · · , h ) MHFLGGMSM(m,p1 ,p2 ,··· ,pn ) hmin n 1 2 1 , h2 , · · · , h n max , · · · , hmax . ≤ MHFLGGMSM(m,p1 ,p2 ,··· ,pn ) hmax , h n 2 1

(4)

Commutativity. Let be any permutation of (h1 , h2 , · · · , hn ), then h10 , h20 , · · · , h0n ( m,p ,p , ··· ,p ) 0 0 0 n 2 1 MHFLGGMSM h1 , h2 , · · · , hn = MHFLGGMSM(m,p1 ,p2 ,··· ,pn ) (h1 , h2 , · · · , hn ).

The proof of property 4 is similar to Property 3, therefore, it is omitted here. Next, by taking into account diverse values of the parameter m, some particular cases of the MHFLGGMSM operator can be discussed as below: (1)

When m = 1, then 1 MHFLGGMSM(1,p1 ) (h1 , h2 , · · · , hn ) = p1 ⊗in=1 ( p1 hi ) n 1 1 n n 1 − 1 = ∪hτ ∈h1 ,hτ ∈h2 ,··· ,hτn ∈hn ϑ p1 ∏k =1 p1 ϑ hτik 2 1 1 n = ∪hτ ∈h1 ,hτ ∈h2 ,··· ,hτn ∈hn ϑ−1 ∏nk=1 ϑ hτ k 2

(2)

When m = 2, then n(n1−1) MHFLGGMSM(2,p1 ,p2 ) (h1 , h2 , · · · , hn ) = p +1 p2 ⊗in=1,j=1,i6= j p1 hi + p2 h j 1 1 n ( n −1) n 1 − 1 = ∪hτ ∈h1 ,hτ ∈h2 ,··· ,hτn ∈hn ϑ p + p2 ∏i =1,j=1,i 6= j p1 ϑ ( hτi ) + p2 ϑ hτj

(38)

(3)

(37)

When m = 3, then MHFLGGMSM(3,p1 ,p2 ,p3 ) (h1 , h2 , · · · , hn ) n(n−11)(n−2) = p1 + p12 + p3 ⊗in=1,j=1,r=1,i6= j6=r p1 hi + p2 h j + p3 hr 1 n(n−1)(n−2) = ∪hτ ∈h1 ,hτ ∈h2 ,··· ,hτn ∈hn ϑ−1 p + p12 + p3 ∏in=1,j=1,r=1,i6= j6=r p1 ϑ(hτi ) + p2 ϑ hτj + p3 ϑ(hτr ) 1

(39)

3.3. WMHFLGMSM Operator Under many practical situations, each attribute has different importance. In this section, taking the weight of the attributes into account, we propose WMHFLGMSM operator which can be defined as below:

Symmetry 2018, 10, 81

14 of 25

Definition 16. Suppose hi (i = 1, 2, · · · , n) is a set of MHFLTESs on S, and w = (w1 , w2 , · · · , wn ) T is the weight vector, which satisfies ∑in=1 wi = 1 and wi > 0(i = 1, 2, · · · , n). Each wi denotes the importance degree of hi . The WMHFLGMSM operator: Ĥ n → Ĥ is: 1 p j  p + p +···+ pm 2 1 m ⊕ ⊗ nw ⊗ h τi j ij j =1  1≤τi1 <···<τim <n  (m,p1 ,p2 ,··· ,pm ) W MHFLMGMSM ( h1 , h2 , · · · , h n ) =   Cnm



(40)

where m = 1, 2, · · · , n and Ĥ is the set containing all MHFLTSs. By the calculation laws for MHFLTSs depicted earlier, we can get the following theorem. Theorem 3. Suppose hi (i = 1, 2, · · · , n) is a set of MHFLTESs on S, and w = (w1 , w2 , · · · , wn ) T is the weight vector, which satisfies ∑in=1 wi = 1 and wi > 0(i = 1, 2, · · · , n). Each wi denotes the importance degree of hi . Then, the overall value aggregated by WMHFLGMSM operator from (40) is still a MHFLTSs and W MHFLGMSM(m,p1 ,p2 ,··· ,pm ) (h1 , h2 , · · · , hn )   Cm ∑k=n 1 ∏m  j=1 nwi k ϑ j  −1  = ∪hτ ∈h1 ,hτ ∈h2 ,··· ,hτn ∈hn ϑ  Cnm  2 1 

!!! p !  j

hτ k ij

1 p1 + p2 +···+ pm

    

  

(41)

where m = 1, 2, · · · , n and k = 1, 2, . . . , Cnm . hτi k represents the i j th element in kth union of each permutation j

which consist of one element from each hi (i = 1, 2, · · · , n). Property 5. (Reducibility). Let w =

1 1 n, n, · · ·

, n1

. Then W MHFLGMSM(m,p1 ,p2 ,··· ,pm ) (h1 , h2 , · · · , hn )

= MHFLGMSM(m,p1 ,p2 ,··· ,pm ) (h1 , h2 , · · · , hn ). Proof

W MHFLGMSM(m,p1 ,p2 ,··· ,pm ) (h1 , h2 , · · · , hn )     −1   ∈h1 ,hτ2 ∈h2 ,··· ,hτn ∈hn  ϑ  1 

Cm ∑k=n 1

∏m j =1

= ∪ hτ

   

= ∪ hτ

∈h1 ,hτ2 ∈h2 ,··· ,hτn ∈hn  ϑ 1

 −1 

 

!!! p !  j

n· n1

ϑ hτ k ij

!! p !  ∏m j =1

ϑ hτ k ij

Cnm

1 p1 + p2 +···+ pm

  

    

  

Cnm

Cm ∑k=n 1

1 p1 + p2 +···+ pm

    

= MHFLGMSM(m,p1 ,p2 ,··· ,pm ) (h1 , h2 , · · · , hn ). 3.4. WMHFLGGMSM Operator In this section, taking the weight into account, we propose WMHFLGGMSM operator which can be defined as follows: Definition 17. Suppose hi (i = 1, 2, · · · , n) is a set of MHFLTESs on S, and w = (w1 , w2 , · · · , wn ) T is the weight vector, which satisfies ∑in=1 wi = 1 and wi > 0(i = 1, 2, · · · , n). wi indicates the weight of hi . The WMHFLGGMSM operator: Ĥ n → Ĥ is:

Symmetry 2018, 10, 81

15 of 25

W MHFLGGMSM(m,p1 ,p2 ,··· ,pm ) (h1 , h2 , · · · , hn ) nwi 1m Cn j 1 m = p + p2 +···+ ⊗ ⊕ p h 1 ≤ τ <···< τ < n j i pm j =1 i im j 1

(42)

where m = 1, 2, · · · , n and Ĥ is the set including all MHFLTSs. By the operations of MHFLTSs depicted in Section 2, we can get the theorem as follows. Theorem 4. Suppose hi (i = 1, 2, · · · , n) is a set of MHFLTESs on S, and w = (w1 , w2 , · · · , wn ) T is the weight vector, which satisfies ∑in=1 wi = 1 and wi > 0(i = 1, 2, · · · , n). wi indicates the weight of hi . Then, the overall value aggregated by WMHFLGGMSM operator from (42) is still a MHFLTSs and 2 ,··· ,pm ) ( h , h , · · · , h ) W MHFLGGMSM(m,p1 ,p n   1 2 nwi 1m Cn m j C m 1 n  = ∪hτ ∈h1 ,hτ ∈h2 ,··· ,hτn ∈hn ϑ−1  p1 + p2 +···+ pm ∏k =1 ∑ j=1 p j ϑ hτi k 2 1

(43)

where m = 1, 2, · · · , n and k = 1, 2, . . . , Cnm . hτi k represents the i j th element in kth union of each permutation j

which consist of one element from each hi (i = 1, 2, · · · , n). Property 6. (Reducibility) T Let w = n1 , n1 , · · · , n1 . Then W MHFLGGMSM(m,p1 ,p2 ,··· ,pm ) (h1 , h2 , · · · , hn ). The proof of property 6 is similar to Property 5, therefore, it is omitted here. 4. A MCDM Approach with MHFLTs In this section, based on the defined aggregation operators, a MCDM approach is developed to process the criteria information with the MHFLTs, which is also valid and feasible for the HFLTS or HFLEs. Considering a MCDM problem as follows. There are d alternatives, denoted by T = {t1 , t2 , · · · , td } and n criteria, denoted by R = {r1 , r2 , · · · , rn }, which weight vector is w = (w1 , w2 , · · · , wn ) T satisfying wi > 0, ∑in=1 wi = 1(i = 1, 2, · · · , n). The assessments are performed by l experts E = {e1 , e2 , · · · , el } and the result of alternative ti given by ek under rj is denoted by aijk based on the LTs S = {si |i = 0, 1, 2, · · · , 2g, g ∈ N }. bij is a MHFLTE that contains inconsecutive and repetitive LTs by combing with the result aijk from l experts, then we obtain the decision matrix B = bij d×n , and the goal is to rank the alternatives. The presented method for this MCDM problem is shown as follows: Step 1. Step 2. Step 3. Step 4.

Normalize the MHFLTE matrix B = bij d×n . Only for the cost criterion rj , bij is normalized by using the negation operator. Integrate the assessment value of every alternative under n criteria and obtain the comprehensive assessment results bi for ti (i = 1, 2, · · · , d). Calculate S(bi ) and V(bi ) for alternative ti . Ranking all alternatives based on S(bi ) and V(bi ).

5. An Illustrative Example In this part, we cited an example (adapted from [10]) to display the flexibility of the proposed methods. In order to get a fast development, the car manufacturer company will select a logistics service provider to concentrate on core competencies. After thorough investigation, five possible

Symmetry 2018, 10, 81

16 of 25

logistic providers will take into account, denoted by T = {t1 , t2 , · · · , t5 }. The criteria, denoted by R = {r1 , r2 , r3 , r4 }, are shown as follows: r1 : cost (such as the total cost of logistics operation); r2 : relationship (such as shared risks and cooperation rewards); r3 : service (such as breadth, specialization, variety); r4 : quality (such as management and improvement). The weight vector of the criteria is w = (0.40, 0.26, 0.17, 0.17) T . The evaluation is carry out by three experts, denoted by E = {e1 , e2 , e3 }. S = {s0 = Extremely Poor (EP), s1 = Very Poor (VP), s2 = Poor (P), s3 = Slightly Poor (SP), s4 = Fair (F), s5 = Slightly Good (SG), s6 = Good (G), s7 = Very Good (VG), s8 = Extremely Good (EG)} be a LTs. The evaluation value of each criteria bij from three experts are represented in the form of MHFLTEs as demonstrated in Table 1. Table 1. Linguistic evaluation result in the form of multi-hesitant linguistic term elements (MHFLTEs). Item

{F, SG, VG}

{SG, SG, EG}

{G, G, EG}

{VP, SP, F}

{SG, G, G}

{G, EG, EG}

{VP, SP, SP }

{SP, G, G}

{P, SP, SP}

{G, G, G}

{P, SP, SG}

{F, F, F}

{VG, VG, VG}

{SP, SG, SG}

{G, EG, EG}

{VP, P, SG}

{G, EG, EG}

{F, SG, SG}

{SP, SG, EG}

{VP, P, P}

The evaluation value in the form of MHFLTSs are transformed into numerical data using Equation (6). If g = 4 and a = 1.4, then ϑ (s0 ) = 0, ϑ (s1 ) = 0.1931, ϑ (s2 ) = 0.3311, ϑ (s3 ) = 0.4296, ϑ (s4 ) = 0.5, ϑ (s5 ) = 0.5704, ϑ (s6 ) = 0.6689, ϑ (s7 ) = 0.8069, ϑ (s8 ) = 1. In order to obtain the best alternative, we adopt the approach described in Section 4 to solve the MCDM problems. 5.1. Procedure of Decision Making Based on WMHFLGMSM Operator Step 1. Get the normalized decision matrix B = bij n×m . Because the four criteria R = {r1 , r2 , r3 , r4 } are all benefit type in this example, thus, this step can be omitted. The decision matrix is shown below:   { s4 , s5 , s7 } { s5 , s5 , s8 } { s6 , s6 , s8 } { s1 , s3 , s4 }  {s , s , s } {s , s , s } {s , s , s } {s , s , s }  6 8 8 3 6 6   5 6 6 1 3 3   B =  { s2 , s3 , s3 } { s6 , s6 , s6 } { s2 , s3 , s5 } { s4 , s4 , s4 }     { s7 , s7 , s7 } { s3 , s5 , s5 } { s6 , s8 , s8 } { s1 , s2 , s5 }  { s6 , s8 , s8 } { s4 , s5 , s5 } { s3 , s5 , s8 } { s1 , s2 , s2 } Step 2.

Step 3.

Integrate the evaluation value of every alternative for four criteria and obtain the comprehensive assessment results bi for ti (i = 1, 2, · · · , 5) by the WMHFGMSM operator (suppose m = 2, p1 = p2 = 1). Calculate S(bi ) for alternative ti (i = 1, 2, · · · , 5), and get S(b1 ) = 0.5970, S(b2 ) = 0.5987, S(b3 ) = 0.4820, S(b4 ) = 0.6329, S(b5 ) = 0.5838.

Step 4.

Rank alternatives based on S(bi ), and get t4 > t2 > t1 > t5 > t3

Symmetry 2018, 10, 81

17 of 25

5.2. Procedure of Decision Making Based on WMHFLGGMSM Operator Step 1. Step 2.

Step 3.

Same as the above step 1. Integrate the evaluation value of every alternative for four criteria and obtain the comprehensive assessment results bi for ti (i = 1, 2, · · · , 5) by the WMHFLGGMSM operator (suppose m = 2, p1 = p2 = 1). Calculate S(bi ) for alternative ti (i = 1, 2, · · · , 5). S(b1 ) = 0.5919, S(b2 ) = 0.5921, S(b3 ) = 0.5109, S(b4 ) = 0.6312, S(b5 ) = 0.5895.

Step 4.

Rank alternatives based on S(bi ), and get t4 > t2 > t1 > t5 > t3

From the above two cases, we can find that the methods based on WMHFLGMSM operator and based on WMHFLGGMSM operator get the same ranking result, and the optimal alternative is all t4 . The WMHFLGMSM operator has the dominant advantage of stressing on the influence of the entire and general data, which permits forceful supplementary among the attribute values, while the WMHFLGGMSM operator has the dominant advantage of stressing on the counterpoise and the coordination among the attribute values. 5.3. Analysis the Effect of the Parameters m, p1 , p2 , . . . , pm In order to observe the effects of the parameters m, p1 , p2 , · · · , pm on this illustrate example, we assign distinct parameter values m, p1 , p2 , · · · , pm to obtain the ranking results, and the results are demonstrated in Tables 2–5. Table 2. Ranking results when m = 2 by weighted generalized MSM operator for MHFLTSs (WMHFLGMSM) operator. p1

Score Function S(bi )

Ranking

S1 = 0.8507, S2 = 0.8497, S3 = 0.5996, S4 = 0.9710, S5 = 0.9769

t5 > t4 > t1 > t2 > t3

S1 = 0.4057, S2 = 0.4194, S3 = 0.3865, S4 = 0.3946, S5 = 0.3209

t2 > t1 > t4 > t3 > t5

S1 = 0.5978, S2 = 0.6047, S3 = 0.5436, S4 = 0.5866, S5 = 0.4691

t2 > t1 > t4 > t3 > t5

S1 = 0.5961, S2 = 0.6371, S3 = 0.4797, S4 = 0.5892, S5 = 0.5371

t2 > t1 > t4 > t3 > t5

S1 = 0.6102, S2 = 0.6720, S3 = 0.4980, S4 = 0.5881, S5 = 0.5394

t2 > t1 > t4 > t3 > t5

S1 = 0.6241, S2 = 0.7026, S3 = 0.5175, S4 = 0.5894, S5 = 0.5442

t2 > t1 > t4 > t3 > t5

S1 = 0.6952, S2 = 0.7033, S3 = 0.5277, S4 = 0.7809, S5 = 0.7526

t4 > t5 > t2 > t1 > t3

S1 = 0.7580, S2 = 0.7672, S3 = 0.5558, S4 = 0.8852, S5 = 0.8705

t4 > t5 > t2 > t1 > t3

S1 = 0.8024, S2 = 0.8102, S3 = 0.5751, S4 = 0.8915, S5 = 0.8758

t4 > t5 > t2 > t1 > t3

S = 0.8356, S2 = 0.8411, S3 = 0.5894, S4 = 0.9106, S5 = 0.9054

t4 > t5 > t2 > t1 > t3

0.5

S1 = 0.5232, S = 0.5295, S3 = 0.4702, S4 = 0.5542, S5 = 0.5010.

t4 > t2 > t1 > t5 > t3

S1 = 0.5970, S2 = 0.5987, S3 = 0.4820, S4 = 0.6329, S5 = 0.5838

t4 > t2 > t1 > t5 > t3

S1 = 0.6619, S2 = 0.6827, S3 = 0.5054, S4 = 0.7118, S5 = 0.6732

t4 > t2 > t1 > t5 > t3

S1 = 0.7010, S2 = 0.7398, S3 = 0.5273, S4 = 0.7575, S5 = 0.7266

t4 > t2 > t1 > t5 > t3

S1 = 0.7285, S2 = 0.7811, S3 = 0.5464, S4 = 0.7879, S5 = 0.7621

t4 > t2 > t1 > t5 > t3

S1 = 0.7872, S2 = 0.8094, S3 = 0.5623, S4 = 0.8112, S5 = 0.7489

t4 > t2 > t1 > t5 > t3

Note: Si is abbreviation of score value S(bi ).

Symmetry 2018, 10, 81

18 of 25

Table 3. Ranking results when m = 2 by WMHFLGGMSM operator. p1

Score Function S(bi )

Ranking

S1 = 0.5990, S2 = 0.4033, S3 = 0.3773, S4 = 0.6969, S5 = 0.6990

t4 > t5 > t1 > t2 > t3

S1 = 0.4537, S2 = 0.4475, S3 = 0.6103, S4 = 0.4372, S5 = 0.3719

t3 > t1 > t2 > t4 > t5

S1 = 0.5191, S2 = 0.5386, S3 = 0.4945, S4 = 0.5286, S5 = 0.4816

t2 > t4 > t1 > t3 > t5

S1 = 0.4690, S2 = 0.4819, S3 = 0.4527, S4 = 0.4799, S5 = 0.4363

t2 > t4 > t1 > t3 > t5

S1 = 0.4327, S2 = 0.4470, S3 = 0.4288, S4 = 0.4389, S5 = 0.3965

t2 > t4 > t1 > t3 > t5

S1 = 0.4084, S2 = 0.4199, S3 = 0.4050, S4 = 0.4092, S5 = 0.3671

t2 > t4 > t1 > t3 > t5

S1 = 0.6741, S2 = 0.6650, S3 = 0.5121, S4 = 0.7551, S5 = 0.7264

t4 > t5 > t1 > t2 > t3

S1 = 0.7461, S3 = 0.7292, S3 = 0.5335, S4 = 0.8521, S5 = 0.8302

t4 > t5 > t1 > t2 > t3

S1 = 0.8075, S2 = 0.7835, S3 = 0.5587, S4 = 0.8604, S5 = 0.9429

t4 > t5 > t1 > t2 > t3

S1 = 0.8604, S2 = 0.8299, S3 = 0.5836, S4 = 0.8960, S5 = 0.8815

t4 > t5 > t1 > t2 > t3

0.5

S1 = 0.6010, S2 = 0.6056, S3 = 0.5523, S4 = 0.6298, S5 = 0.5787

t4 > t2 > t1 > t5 > t3

S1 = 0.5919, S2 = 0.5921, S3 = 0.5109, S4 = 0.6312, S5 = 0.5895

t4 > t2 > t1 > t5 > t3

S1 = 0.6063, S2 = 0.6240, S3 = 0.4903, S4 = 0.6580, S5 = 0.6016

t4 > t2 > t1 > t5 > t3

S1 = 0.6261, S2 = 0.6554, S3 = 0.4882, S4 = 0.6857, S5 = 0.6184

t4 > t2 > t1 > t5 > t3

S1 = 0.6454, S2 = 0.6827, S3 = 0.4911, S4 = 0.7108, S5 = 0.6354

t4 > t2 > t1 > t5 > t3

S1 = 0.6633, S2 = 0.7068, S3 = 0.4960, S4 = 0.7334, S5 = 0.6515

t4 > t2 > t1 > t5 > t3

Note: Si is abbreviation of score value S(bi ).

Table 4. Ranking results when m = 3 by WMHFLGMSM operator. p1

Score Function S(bi )

Ranking

S1 = 0.4856, S2 = 0.4956, S3 = 0.4583, S4 = 0.5140, S5 = 0.4644

t4 > t2 > t1 > t5 > t3

S1 = 0.9580, S2 = 0.9464, S3 = 0.6501, S4 = 0.9609, S5 = 0.9804

t5 > t4 > t2 > t1 > t3

S1 = 0.6360, S2 = 0.5600, S3 = 0.4987, S4 = 0.5747, S5 = 0.5111

t1 > t4 > t2 > t5 > t3

S1 = 0.2905, S2 = 0.3492, S3 = 0.3304, S4 = 0.3054, S5 = 0.2259

t2 > t3 > t4 > t1 > t5

S1 = 0.5251, S2 = 0.4998, S3 = 0.4302, S4 = 0.5567, S5 = 0.4791

t4 > t1 > t2 > t5 > t3

S1 = 0.5119, S2 = 0.4736, S3 = 0.4088, S4 = 0.5513, S5 = 0.4618

t4 > t1 > t2 > t5 > t3

S1 = 0.5069, S2 = 0.4581, S3 = 0.3958, S4 = 0.5527, S5 = 0.4564

t4 > t1 > t2 > t5 > t3

S1 = 0.5056, S2 = 0.4480, S3 = 0.3872, S4 = 0.5561, S5 = 0.4514

t4 > t1 > t2 > t5 > t3

S1 = 0.5885, S2 = 0.6040, S3 = 0.4933, S4 = 0.5765, S5 = 0.5302

t2 > t1 > t4 > t5 > t3

S1 = 0.6106, S2 = 0.6506, S3 = 0.5181, S4 = 0.5739, S5 = 0.5356

t2 > t1 > t4 > t5 > t3

S1 = 0.6278, S2 = 0.6857, S3 = 0.5398, S4 = 0.5732, S5 = 0.5406

t2 > t1 > t4 > t5 > t3

S1 = 0.6415, S2 = 0.7165, S3 = 0.5579, S4 = 0.5734, S5 = 0.5450

t2 > t1 > t4 > t5 > t3

S1 = 0.6449, S2 = 0.6395, S3 = 0.5089, S4 = 0.7142, S5 = 0.6661

t4 > t5 > t1 > t2 > t3

S1 = 0.7052, S2 = 0.7004, S3 = 0.5341, S4 = 0.8110, S5 = 0.7721

t4 > t5 > t1 > t2 > t3

S1 = 0.7498, S2 = 0.7447, S3 = 0.5521, S4 = 0.8844, S5 = 0.8534

t4 > t5 > t1 > t2 > t3

S1 = 0.7841, S2 = 0.7783, S3 = 0.5658, S4 = 0.9416, S5 = 0.9173

t4 > t5 > t1 > t2 > t3

Note: Si is abbreviation of score value S(bi ).

Symmetry 2018, 10, 81

19 of 25

Table 5. Ranking results when m = 3 by WMHFLGGMSM operator. p1

Score Function S(bi )

Ranking

S1 = 0.6215, S2 = 0.6304, S3 = 0.5656, S4 = 0.6505, S5 = 0.5933

t4 > t2 > t1 > t5 > t3

S1 = 0.5398, S2 = 0.5622, S3 = 0.2976, S4 = 0.7049, S5 = 0.7455

t5 > t4 > t2 > t1 > t3

S1 = 0.7604, S2 = 0.4687, S3 = 0.6121, S4 = 0.6823, S5 = 0.6272

t1 > t4 > t5 > t3 > t2

S = 0.4309, S = 0.4387, S3 = 0.6095, S4 = 0.4303, S5 = 0.3510

t3 > t2 > t1 > t4 > t5

S = 0.5274, S2 = 0.5573, S3 = 0.4807, S4 = 0.5451, S5 = 0.5126

t2 > t4 > t3 > t1 > t5

S1 = 0.4703, S2 = 0.4968, S3 = 0.4468, S4 = 0.4974, S5 = 0.4501

t2 > t4 > t3 > t1 > t5

S1 = 0.4276, S2 = 0.4589, S3 = 0.4188, S4 = 0.4483, S5 = 0.4046

t2 > t4 > t3 > t1 > t5

S1 = 0.3946, S2 = 0.4286, S3 = 0.3957, S4 = 0.4126, S5 = 0.3699

t2 > t4 > t3 > t1 > t5

S1 = 0.6105, S2 = 0.5878, S3 = 0.5184, S4 = 0.6209, S5 = 0.5792

t1 > t4 > t2 > t5 > t3

S1 = 0.6045, S2 = 0.5676, S3 = 0.5116, S4 = 0.5987, S = 0.5579

t1 > t4 > t2 > t5 > t3

S1 = 0.5962, S2 = 0.5489, S3 = 0.5032, S4 = 0.5802, S5 = 0.5402

t1 > t4 > t2 > t5 > t3

S1 = 0.5876, S2 = 0.5322, S3 = 0.4947, S4 = 0.5647, S5 = 0.5255

t1 > t4 > t2 > t5 > t3

S1 = 0.6931, S2 = 0.6984, S3 = 0.5173, S4 = 0.7903, S5 = 0.8062

t5 > t4 > t2 > t1 > t3

S1 = 0.7968, S2 = 0.8078, S3 = 0.5444, S4 = 0.8108, S5 = 0.8273

t5 > t4 > t2 > t1 > t3

S1 = 0.90422, S2 = 0.9215, S3 = 0.5935, S4 = 0.9266, S5 = 0.9329

t5 > t4 > t2 > t1 > t3

S1 = 0.9275, S2 = 0.9297, S3 = 0.6308, S4 = 0.9439, S5 = 0.9527

t5 > t4 > t2 > t1 > t3

Note: Si is abbreviation of score value S(bi ).

As we can see from Tables 2–5, the ranking results may be different for the distinct parameter values. (1)

(2)

(3)

If we keep the balance of parameter assignment (i.e., p1 = p2 = · · · = pm ), the ranking results of alternatives are identical in the condition of distinct parameter values. If the balance of parameter assignment is broken, the ranking results of alternatives will start to be changed, but as the value of parameter gap becomes lager and lager or the value of one parameter is much larger than other parameters, the ranking results will not be changed. So, the risk preference of experts plays an important role in real MCDM. In Tables 2 and 3, if p1 = 0 or p2 = 0, the WMHFLGMSM operator and WMHFLGGMSM operator cannot consider the interrelation of the criteria. Moreover, the similar situation exists in Tables 4 and 5. For the WMHFLGMSM operator, the greater value of m becomes, in other words, the more interrelationships of criteria we take into account, the smaller value of score function will get. Nevertheless, for the WMHFLGGMSM operator, the results are the opposite, the greater value of m becomes, in other words, the more interrelationships of criteria we take into account, the greater value of score function will get.

Therefore, experts should select the suitable parameter values in order to keep their risk preference at real MCDM environment. If expert has risk preference, he/she should assign the parameter as large as possible; if expert has risk aversion, he/she should assign the parameter as small as possible. Generally, we propose that experts assign the values of the parameters as m = 2, p1 = p2 = 1 in practical problems, which are not only simple for operations, but also take into account the interrelations for two parameters. 5.4. Comparison with the Other Methods To verify the validity of the proposed approaches, we use the existing generalized weighted average operator for HFLSs (GHFLWA) [45], weighted BM operator for HFLSs (WHFLBM) [33]

Symmetry 2018, 10, 81

20 of 25

and multi-HFLS weighted generalized HM (MHFL-WGHM) operator [10] respectively to solve an illustrative example from [40] which is listed as follows: For the sake of further optimizing healthcare resource allocation, we need to select a general hospital to improve the traditional healthcare system and build a new comprehensive healthcare system. After thorough investigation, four general hospitals will take into account, denoted by T = {t1 , t2 , t3 , t4 }, namely SU hospital (t1 ), FU hospital (t2 ), UMC hospital (t3 ), PLA hospital (t4 ). Three main criteria, denoted by R = {r1 , r2 , r3 }, are shown as follows: service environment (r1 ), diagnosis and treatment (r2 ) and social resource allocation (r3 ); The weight vector of the criteria is w = (0.3, 0.2, 0.5) T . The linguistic evaluation result is listed in Table 6 and the ranking results of different methods are shown in Table 7. Table 6. The Linguistic evaluation result for instance proposed by [40]. Item

{ s3 , s4 }

{ s5 }

{ s4 } { s4 } { s5 }

{ s2 , s3 } { s4 , s5 } { s3 , s4 }

{ s5 } { s5 } { s5 , s6 } { s4 }

t3 t4

Table 7. The comparisons of different methods for example proposed by [40]. Aggregation Operator

Parameter Value

Ranking Results

GHFLWA [45]

Îť=1

t3 > t2 > t4 > t1

p=q=1

t3 > t4 > t2 > t1

p=q=1

t3 > t4 > t2 > t1

W HFLBM

( p,q)

[33]

MHFL-WGHM( p,q) W MHFLGMSM(1,p)

[10]

proposed in this paper

W MHFLGMSM(2,p1, p2 )

proposed in this paper

W MHFLGMSM(3,p1, p2 ,p3 ) proposed in this paper

m = 1, p = 1

t3 > t2 > t4 > t1

m = 2, p1 =p2 = 1

t3 > t4 > t2 > t1

m = 3, p1 = p2 = p3 = 1

t3 > t4 > t2 > t1

From Table 7, we can find that the approach based on the W MHFLGMSM(1,p) operator in this paper and the approach based on GHFLWA operator [45] have the same ranking result, i.e., t3 > t2 > t4 > t1 , and the method based on W MHFLGMSM(2,p1, p2 ) operator proposed in this paper and the method based on WHFLBM operator and MHFL-WGHM operator have also the same ranking result, i.e., t3 > t4 > t2 > t1 . Obviously, we can easily explain these results (1) because both of the proposed approaches based on the W MHFLGMSM(1,p) operator in this paper and the approach based on GHFLWA operator [45] donâ€™t consider the interrelationship of criteria, they produce the same ranking result; (2) while all of the proposed methods based on the W MHFLGMSM(2,p1, p2 ) operator and the approach based on WHFLBM operator [33] and MHFL-WGHM operator [10] can consider the interrelationship between two criteria, they get the same ranking result. So, these results can demonstrate that the presented approaches are rational and valid in HFLS. In order to show the advantage of the novel method based on WMHFLGMSM operator in this paper, we continue to calculate the same example from [10] by comparing with some existing methods. We can get the final result demonstrated in Table 8. From Table 8, we can find that the final result by the approach based on the MHFL-WGHM operator [10] is the same with using the WMHFLGMSM operator proposed in this paper, whereas the methods using GHFLWA operator [45] and WHFLBM operator [33] and the method using the WMHFLGMSM operator get different ranking results.

Symmetry 2018, 10, 81

21 of 25

Table 8. The comparisons of different methods. Aggregation Operator

Parameter Value

Ranking Results

GHFLWA [45]

位=1

t5 > t4 > t1 > t2 > t3

m = 2, p1 = p2 = 1

t1 > t4 > t5 > t2 > t3

p=q=1

t4 > t2 > t1 > t5 > t3

m = 2, p1 = p2 = 1

t4 > t2 > t1 > t5 > t3

W HFLBM

( p,q)

[33]

MHFL-WGHM( p,q) W MHFLGMSM(m,p1, p2 ,路路路 pn )

[10]

prposed in this paper

In the following, we analyze the reason of producing these results as follows. (1)

(2)

The final overall results from the GHFLWA operator [45] and the WHFLBM operator [33] are achieved under the condition that the repeated assessment values in Table 1 are eliminated. Obviously, this can lead to information loss, and make the ranking results unreasonable. From Table 8, we can also get this conclusion because they produced the different ranking results. The approaches based on the MHFL-WGHM operator [10] and the WMHFLGMSM operator proposed in this paper produced the same ranking results because they can fully express the evaluation information, and all considered the interrelationship between two criteria. However, the proposed method based on the WMHFLGMSM operator in this paper can consider the interrelationship among any number of attributes by some parameters.

Obviously, the presented approach based on the WMHFLGMSM operator in this paper can get over the weakness of the approaches based on the GHFLWA operator [45] and the WHFLBM operator [33] because the HFLTSs cannot sufficiently convey the hesitance of experts, and WMHFLGMSM operator in this paper can take into the interrelation among any number of the criteria, which is more general than some existing methods. In summary, the presented approach based on the WMHFLGMSM operator not only considers the significance of repeated assessment values, but also takes into account the interrelationships among any criteria. In the following, we give a detailed contrastive analysis for the different methods, and state these clearly in Table 9. Table 9. The comparisons of the features for different methods. Describe Fuzzy Information and Easier Parameters

Operators

GHFLWA [45]

Consider the Interrelationship of Two Arguments

Consider the Interrelationship of Any Multi-Input Arguments

Consider the Importance of Repeated Linguistic Values

Yes

MHFL-WGHM( p,q) [10]

Yes

The operators prposed in this paper

Yes

W HFLBM

( p,q)

[33]

From Table 9, we obtain the conclusion as follows: (1)

Compared with the approach based on GHFLWA operator proposed by [45], we can note the weakness of the GHFLWA operator is that the input assessment values are independent and does not think about the interrelationships among input arguments. Under these circumstances, the GHFLWA operator proposed by [45] is a particular example of WMHFLGMSM operator when m = 1, p = 1. Our novel method takes into account the real decision environment in which there are some relationships among some criteria. In above illustrative example, the level of management and improvement will have an effect on the breadth, specialization, variety of logistic company, that is to say we should consider the interrelationships between r3 and r4 . Therefore, our presented approach is more suitable for dealing with actual decision problems

Symmetry 2018, 10, 81

(2)

(3)

22 of 25

than the presented approach based on GHFLWA operator proposed by [45]. Another weakness of the presented approach based on GHFLWA operator [45] is that it adopts HFLTSs which cannot sufficiently convey the hesitance of experts. At the same time, our presented approach can completely express the evaluation information. Compared with the approach based on the WHFLBM operator presented by [33], our new method can preserve the repetition of linguistic evaluation information and considers the interrelationship among more than two input arguments. However, the approach based on the WHFLBM operator only considers the interrelationship between two input arguments and the frequencies of repeated values are neglected. Therefore, as an extension of HFLSs and HFLTSs, the presented approach based on the WMHFLGMSM operator is more reasonable to aggregate the repeated linguistic information in practice. Moreover, we also noticed that the WMHFLGMSM operator with parameters will degrade into the WHFLBM operator proposed by [33] if m = 2, p1 = 1, p2 = 1. Therefore, the WMHFLGMSM operators have more generality and are more robust. Compared with the approach based on the MHFL-WGHM operator presented by [10], our presented novel method can take into account the correlation among multi-inputs. However, the MHFL-WGHM operator can only consider correlation between two inputs. In many real decision-making problems, the interrelationships among multi-input arguments must be considered. Thus, the presented approach is more general and wider, and it is more adequate to deal with MCDM problems.

By the contrast analysis above, we can show that the WMHFLGMSM operator we prosed in this paper is better than the WHFLBM operator proposed by [33], the MHFL-WGHM operator proposed by [10], and the GHFLWA operator proposed by [45] for aggregating the multi-hesitant fuzzy linguistic information. 6. Conclusions A MHFLTS is an extension of HFLS and HFLTS, which can preserve the frequencies of repeated LTs. In addition, the MSM operator can take into account the correlativity among multi-inputs. In this paper, we investigate the MCDM problems with the information of the MHFLTSs. Firstly, we developed the MHFLGMSM operator, MHFLGGMSM operator, WMHFLGMSM operator, and WMHFLGGMSM operator, respectively. Then, their properties and particular cases of these operators are further studied. Based on these operators, we proposed a method to solve the MCDM problem with MHFLTEs. Finally, some examples are provided to contrast the presented approaches with the existing approaches, such the methods based on the WHFLBM operator [33], the MHFL-WGHM operator [10], the GHFLWA operator [45]. The comparison results demonstrate that the presented approaches outperform some existing approaches [10,33,45]. Fortunately, we maybe use these novel operators to solve new MCDM problems in practice, such as industrial site selection [46], intrusion detection [47], EEG signals analysis [48], consensus model [49], temporary disaster debris Management [50], etc. The limitation of proposed operators is that if we consider interrelationships among more than three criteria, more parameters will be given, which may bewilder experts, such that they perhaps cannot decide how to choose the values of parameters. In further research, it is necessary to apply these operators to real decision-making problems, such as evaluations on population resources and environment [51â€“54] or Chinese culture [55]. Acknowledgments: This paper is supported by the National Natural Science Foundation of China (Nos. 71771140 and 71471172), the Special Funds of Taishan Scholars Project of Shandong Province (No. ts201511045), Shandong Provincial Social Science Planning Project (Nos. 17BGLJ04, 16CGLJ31 and 16CKJJ27), the Natural Science Foundation of Shandong Province (No. ZR2017MG007), Key research and development program of Shandong Province (No. 2016GNC110016), and the Science Research Foundation of Heze University (No. XY16SK02).

Symmetry 2018, 10, 81

23 of 25

Author Contributions: Peide Liu gave the research ideas, proposed a research method, revised whole manuscript, and responded to reviewers’ comments; Hui Gao designed the experiments, performed the experiments, analyzed the results, and wrote whole manuscript. Conflicts of Interest: The authors declare no conflict of interest.

References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

11. 12. 13.

14. 15. 16. 17. 18. 19. 20. 21. 22.

Zadeh, L.A. The concept of a linguistic variable and its application to approximate reasoning. Inf. Sci. 1975, 8, 199–249. [CrossRef] Chang, S.L.; Wang, R.C.; Wang, S.Y. Applying fuzzy linguistic quantifier to select supply chain partners at different phases of product life cycle. Int. J. Prod. Econ. 2006, 100, 348–359. [CrossRef] Yan, H.B.; Ma, T.; Li, Y. A novel fuzzy linguistic model for priority engineering design requirements in quality function deployment under uncertainties. Int. J. Prod. Res. 2013, 51, 6336–6355. [CrossRef] Lin, Q.L.; Liu, L.; Liu, H.C. Integrating hierarchical balanced scorecard with fuzzy linguistic for evaluating operating room performance in hospitals. Expert Syst. Appl. 2013, 40, 1917–1924. [CrossRef] Montes, R.; Villar, P.; Herrer, F. A web tool to support decision making in the housing market using hesitant fuzzy linguistic term sets. Appl. Soft Comput. 2015, 35, 949–957. [CrossRef] Fattahi, R.; Khalilzadeh, M. Risk evaluation using a novel hybrid method based on FMEA, extended MULTIMOORA, and AHP methods under fuzzy environment. Saf. Sci. 2018, 102, 290–300. [CrossRef] Rodriguez, R.M.; Martinez, L.; Herrera, F. Hesitant Fuzzy Linguistic Term Sets for Decision Making. IEEE Trans. Fuzzy Syst. 2012, 20, 109–119. [CrossRef] Rodriguez, R.M.; Martinez, L.; Herrera, F. A group decision making model dealing with comparative linguistic expressions based on hesitant fuzzy linguistic term sets. Inf. Sci. 2013, 241, 28–42. [CrossRef] Liu, P.; Chen, S.M. Multiattribute group decision making based on intuitionistic 2-tuple linguistic information. Inf. Sci. 2018, 430–431, 599–619. [CrossRef] Wang, J.; Wang, J.Q.; Tian, Z.; Zhao, D. A multi-hesitant fuzzy linguistic multicriteria decision-making approach for logistics outsourcing with incomplete weight information. Int. Trans. Oper. Res. 2018, 25, 831–856. [CrossRef] Wang, J.; Wang, J.Q.; Zhang, H.Y. A likelihood-based TODIM approach based on multi-hesitant fuzzy linguistic information for evaluation in logistics outsourcing. Comput. Ind. Eng. 2016, 99, 287–299. [CrossRef] Liu, P.; Su, Y. The Extended TOPSIS Based on Trapezoid Fuzzy Linguistic Variables. J. Converg. Inf. Technol. 2010, 5, 38–53. Li, C.C.; Rodrıguez, R.M.; Martınez, L.; Dong, Y.; Herrera, F. Personalized individual semantics based on consistency in hesitant linguistic group decision making with comparative linguistic expressions. Knowl.-Based Syst. 2018, 145, 156–165. [CrossRef] Liao, H.; Xu, Z.; Zeng, X.J. Distance and similarity measures for hesitant fuzzy linguistic term sets and their application in multi-criteria decision making. Inf. Sci. 2014, 271, 125–142. [CrossRef] Chen, S.M.; Hong, J.A. Multicriteria linguistic decision making based on hesitant fuzzy linguistic term sets and the aggregation of fuzzy sets. Inf. Sci. 2014, 286, 63–74. [CrossRef] Wang, J.Q.; Wang, J.; Chen, Q.H. An outranking approach for multi-criteria decision-making with hesitant fuzzy linguistic term sets. Inf. Sci. 2014, 280, 338–351. [CrossRef] Zhu, B.; Xu, Z. Consistency Measures for Hesitant Fuzzy Linguistic Preference Relations. IEEE Trans. Fuzzy Syst. 2014, 22, 35–45. [CrossRef] Zhang, Z.; Wu, C. On the use of multiplicative consistency in hesitant fuzzy linguistic preference relations. Knowl.-Based Syst. 2014, 72, 13–27. [CrossRef] Tong, X.; Yu, L. MADM Based on distance and correlation coefficient measures with decision-maker preferences under a hesitant fuzzy environment. Soft Comput. 2016, 20, 4449–4461. [CrossRef] Beg, I.; Rashid, T. TOPSIS for Hesitant Fuzzy Linguistic Term Sets. Int. J. Intell. Syst. 2013, 28, 1162–1171. [CrossRef] Rashid, T.; Faizi, S.; Xu, Z.; Zafar, S. ELECTRE-Based Outranking Method for Multi-criteria Decision Making Using Hesitant Intuitionistic Fuzzy Linguistic Term Sets. Int. J. Fuzzy Syst. 2018, 20, 78–92. [CrossRef] Liao, H.; Xu, Z.; Zeng, X.J. Hesitant Fuzzy Linguistic VIKOR Method and Its Application in Qualitative Multiple Criteria Decision Making. IEEE Trans. Fuzzy Syst. 2015, 23, 1343–1355. [CrossRef]

Symmetry 2018, 10, 81

23. 24. 25. 26. 27.

28. 29. 30.

31.

32.

33. 34. 35. 36. 37. 38. 39. 40. 41. 42. 43. 44. 45. 46.

24 of 25

Yu, W.; Zhang, Z.; Zhong, Q. Extended TODIM for multi-criteria group decision making based on unbalanced hesitant fuzzy linguistic term sets. Comput. Ind. Eng. 2017, 114, 316–328. [CrossRef] Farhadinia, B.; Herrera, E. Entropy Measures for Hesitant Fuzzy Linguistic Term Sets Using the Concept of Interval-Transformed Hesitant Fuzzy Elements. Int. J. Fuzzy Syst. 2017. [CrossRef] Zang, Y.; Sun, W.; Han, S. Grey relational projection method for multiple attribute decision making with interval-valued dual hesitant fuzzy information. J. Intell. Fuzzy Syst. 2017, 33, 1053–1066. [CrossRef] Xu, Z.S.; Pan, L.; Liao, H.C. Multi-criteria decision-making method of hesitant fuzzy linguistic term set based on improved MACBETH method. Control Decis. 2017, 32, 1266–1272. Xu, Y.; Xu, A.; Wang, H. Hesitant fuzzy linguistic linear programming technique for multidimensional analysis of preference for multi-attribute group decision making. Int. J. Mach. Learn. Cybern. 2016, 7, 845–855. [CrossRef] Zhang, W.; Ju, Y.; Liu, X. Multiple criteria decision analysis based on Shapley fuzzy measures and interval-valued hesitant fuzzy linguistic numbers. Comput. Ind. Eng. 2017, 105, 28–38. [CrossRef] Porcel, C.; Ching-López, A. Fuzzy linguistic recommender systems for the selective diffusion of information in digital libraries. J. Inf. Process. Syst. 2017, 13, 653–667. [CrossRef] Sha, J.; Zhang, Y. Fuzzy Petri Net Method for Hesitant Decision Making. In Proceedings of the 2016 International IEEE Conferences Ubiquitous Intelligence & Computing, Scalable Computing and Communications, Advanced and Trusted Computing, Cloud and Big Data Computing, Internet of People, and Smart World Congress, Toulouse, France, 18–21 July 2016; pp. 942–948. Wang, J.; Wang, J.Q.; Zhang, H.Y. Multi-criteria Group Decision-Making Approach Based on 2-Tuple Linguistic Aggregation Operators with Multi-hesitant Fuzzy Linguistic Information. Int. J. Fuzzy Syst. 2016, 18, 81–97. [CrossRef] Wu, Z. Aggregation Operators for Hesitant Fuzzy Linguistic Term Sets with an Application in MADM. In Proceedings of the 27th Chinese Control and Decision Conference, Qingdao, China, 23–25 May 2015; pp. 879–884. Gou, X.; Xu, Z.; Liao, H. Multiple criteria decision making based on Bonferroni means with hesitant fuzzy linguistic information. Soft Comput. 2017, 21, 6515–6529. [CrossRef] Zhu, C.; Zhu, L.; Zhang, X. Linguistic hesitant fuzzy power aggregation operators and their applications in multiple attribute decision-making. Inf. Sci. 2016, 367, 809–826. [CrossRef] Wang, J.Q.; Peng, L.; Zhang, H.Y. Method of multi-criteria group decision-making based on cloud aggregation operators with linguistic information. Inf. Sci. 2014, 274, 177–191. [CrossRef] Liu, P.; Mahmood, T.; Khan, Q. Multi-Attribute Decision-Making Based on Prioritized Aggregation Operator under Hesitant Intuitionistic Fuzzy Linguistic Environment. Symmetry 2017, 9, 270. [CrossRef] Maclaurin, C. A second letter to Martin Folkes, Esq.: Concerning the roots of equations, with the demonstration of other rules of algebra. Philos. Trans. R. Soc. Lond. Ser. A 1729, 36, 59–96. Detemple, D.; Robertson, J. On generalized symmetric means of two variables. Angew. Chem. 1979, 47, 4638–4660. Liu, P.; Zhang, X. Some Maclaurin Symmetric Mean Operators for Single-Valued Trapezoidal Neutrosophic Numbers and Their Applications to Group Decision Making. Int. J. Fuzzy Syst. 2018, 20, 45–61. [CrossRef] Liu, P.; Qin, X. Maclaurin symmetric mean operators of linguistic intuitionistic fuzzy numbers and their application to multiple-attribute decision-making. J. Exp. Theor. Artif. Intell. 2017, 29, 1173–1202. [CrossRef] Qin, J.; Liu, X. Approaches to uncertain linguistic multiple attribute decision making based on dual Maclaurin symmetric mean. J. Intell. Fuzzy Syst. 2015, 29, 171–186. [CrossRef] Zhang, Z.; Wei, F. Approaches to comprehensive evaluation with 2-tuple linguistic information. J. Intell. Fuzzy Syst. 2015, 28, 469–475. Wang, J.Q.; Yang, Y.; Li, L. Multi-criteria decision-making method based on single-valued neutrosophic linguistic Maclaurin symmetric mean operators. Neural Comput. Appl. 2016. [CrossRef] Liu, P.; Teng, F. Multiple attribute group decision making methods based on some normal neutrosophic number Heronian Mean operators. J. Intell. Fuzzy Syst. 2017, 32, 2375–2391. [CrossRef] Zhang, Z.; Wu, C. Hesitant fuzzy linguistic aggregation operators and their applications to multiple attribute group decision making. J. Intell. Fuzzy Syst. 2014, 26, 2185–2202. Taibi, A.; Atmani, B. Combining fuzzy AHP with GIS and decision rules for industrial site selection. Int. J. Interact. Multimed. Artif. Intell. 2017, 4, 60–69. [CrossRef]

Symmetry 2018, 10, 81

47. 48. 49.

50. 51. 52. 53. 54. 55.

25 of 25

Harish, B.S.; Kumar, S.V.A. Anomaly based intrusion detection using modified fuzzy clustering. Int. J. Interact. Multimed. Artif. Intell. 2017, 4, 54–59. [CrossRef] Esmaeilpour, M.; Mohammadi, A.R.A. Analyzing the EEG signals in order to estimate the depth of anesthesia using wavelet and fuzzy neural networks. Int. J. Interact. Multimed. Artif. Intell. 2017, 4, 12–15. [CrossRef] Cabrerizo, F.J.; Morente-Molinera, J.A.; Pedrycz, W.; Taghavi, A.; Herrera-Viedma, E. Granulating linguistic information in decision making under consensus and consistency. Expert Syst. Appl. 2018, 99, 83–92. [CrossRef] Habib, M.S.; Sarkar, B. An Integrated Location-Allocation Model for Temporary Disaster Debris Management under an Uncertain Environment. Sustainability 2017, 9, 716. [CrossRef] Zha, D.; Kavuri, A.S. Effects of technical and allocative inefficiencies on energy and nonenergy elasticities: an analysis of energy-intensive industries in China. Chin. J. Popul. Resour. Environ. 2016, 14, 292–297. [CrossRef] Zhang, B.; He, M.; Pan, H. A study on the design of a hybrid policy for carbon abatement. Chin. J. Popul. Resour. Environ. 2017, 15, 50–57. [CrossRef] Zhang, R.; Shi, G. Analysis of the relationship between environmental policies and air quality during major social events. Chin. J. Popul. Resour. Environ. 2016, 14, 167–173. [CrossRef] Zhu, J. The 2030 Agenda for Sustainable Development and China’s implementation. Chin. J. Popul. Resour. Environ. 2017, 15, 142–146. [CrossRef] Hou, Y.S. Reflection on the inheritance and development of “Wang PI opera” in Shandong. Dong Yue Trib. 2016, 37, 142–146. (In Chinese) © 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).