You are on page 1of 9

Engineering Applications of Artificial Intelligence 104 (2021) 104395

Contents lists available at ScienceDirect

Engineering Applications of Artificial Intelligence


journal homepage: www.elsevier.com/locate/engappai

Picture fuzzy set and quality function deployment approach based novel
framework for multi-criteria group decision making method
Akanksha Singh, Sanjay Kumar ∗
Department of Mathematics, Statistics and Computer Science, G. B. Pant University of Agriculture and Technology, Pantnagar, Uttarakhand, 263145, India

ARTICLE INFO ABSTRACT


Keywords: Quality function deployment (QFD) translates customers’ requirements into product design requirements for
Picture fuzzy set improving quality and design of the product. Very often, customers’ requirements are uncertain because of
Quality function deployment imprecision and linguistic representation. Many researchers have integrated fuzzy set (FS) and intuitionistic
House of quality
fuzzy set (IFS) with QFD to model uncertainty in the preferences of customers’ requirements due to imprecision
Multi-criteria group decision making
and vagueness. Since picture fuzzy set (PFS) which is an extension of IFS, has been proved more ideal than
Social media platforms
fuzzy and intuitionistic fuzzy set in modeling of fuzziness and uncertainty in real life problem of decision
making, we have integrated QFD with PFS to propose a multi-criteria group decision making (MCGDM) method.
Present study contributes a MCGDM method that uses prominent characteristic of QFD in the integration of
heterogeneous preferences of decision makers by splitting them into groups of users and analysts. Positive,
negative and no correlations among engineering criteria without normalization in QFD is used in proposed
MCGDM method to obtain more precise overall relationship matrix and hence final ranking of alternatives.
Feasibility and suitability of proposed MCGDM method is verified by using it on a real case study of ranking
of Facebook, Instagram, WhatsApp and Twitter in which users are taken as decision makers.

1. Introduction Terninko, 2018; Chen et al., 2016; de Fátima Cardoso et al., 2015;
Abdul-Rahman et al., 1999; Dikmen et al., 2005; Barnett and Raja,
Multi-criteria group decision making (MCGDM) is a process of 1995; Herzwurm et al., 2015; Quinn et al., 2009; Zhong et al., 2018;
evaluating a finite set of alternatives by a group of decision makers Onar et al., 2016; Wu et al., 2018; Lee et al., 2017) have pro-
against certain criteria. Due to wide applicability of MCGDM in real life, posed its applications in product design, development of an organic
its profound applications are found in human resource, teacher’s per- product, construction design, software development process, higher
formance evaluation, stock selection, supplier selections and in many education, computer workstation selection, hospitality industry and
more. Decision making with uncertain, imprecise, vague information supplier selection.
is itself a complex problem and it becomes more complex when the
Although QFD has been used extensively for designing a product
alternatives need to be assessed with respect to large number of criteria.
according to needs of the customers, it can also be explored in decision
Decision making where customers’ satisfaction is at utmost priority, one
making problems where alternatives are to be ranked using the criteria
level evaluation either by set of experts or set of users do not help in
set by technical expert and end users. Various inputs in the form
achieving their all objectives. MCGDM methods that evaluate products
need to include both customers’ voice and experts ‘analysis. of judgment and evaluation are required for qualitative analysis in
Quality function deployment (QFD) was developed in late 1960 QFD process. Very often inputs required in QFD are represented using
in Japan while working for Mitsubishi’s shipyard. According to Akao linguistic terms and exhibits uncertainties that cannot be dealt by
(1994), QFD is a structured, user-driven quality assurance and prod- probability theory.
uct planning tool that transforms qualitative customers’ demand into Zadeh (1965) defined fuzzy sets which have been established as
quantitative parameters to incorporate customers’ voice into technical an ideal tool to model uncertainty in a system due to aforesaid non
characteristics. . The basic idea behind QFD is to understand customers’ stochastic reasons. Earlier, Lee (1990a,b) used fuzzy sets for convert-
needs and transform them into measurable technical specifications. ing linguistic control strategy using fuzzy control rules and examined
QFD produces more accurate decisions by focusing on several aspects decision making logic of fuzzy logic controller. Chen (1996) devel-
and criteria based on customers needs. In view of features and ca- oped a weighted fuzzy reasoning for fuzzy rule-based system. Chen
pabilities of QFD, many researchers (Sireli et al., 2007; Sakao, 2007; and Jong (1997) and Chen and Huang (2003) explored fuzzy sets in

∗ Corresponding author.
E-mail addresses: singhakanksha981993@gmail.com (A. Singh), skruhela@hotmail.com (S. Kumar).

https://doi.org/10.1016/j.engappai.2021.104395
Received 27 January 2021; Received in revised form 11 July 2021; Accepted 14 July 2021
Available online 28 July 2021
0952-1976/© 2021 Elsevier Ltd. All rights reserved.
A. Singh and S. Kumar Engineering Applications of Artificial Intelligence 104 (2021) 104395

relational database management system. Aslan et al. (2017) proposed Proposed method not only includes inherent characteristics of QFD
fuzzy PID controller for blast furnace top gas pressure. Song and but also provides an approach to deal all possible types of the corre-
Chissom (1993a,b) explored fuzzy sets in time series forecasting and lations among ECs to obtain more precise overall relationship matrix
later many researchers (Huarng and Yu, 2006; Kuo et al., 2009; Cai with picture fuzzy information. Since problem of selection of social
et al., 2013; Chen and Kao, 2013; Altan and Karasu, 2019; Karasu media platforms depends upon the number of criteria and has been a
et al., 2020) integrated evolutionary computing techniques of neural daunting task, feasibility, and efficiency of proposed MCGDM method
network, genetic algorithm, particle swarm optimization and support is exhibited by implementing it for the selection of social networking
vector machine with fuzzy sets to develop forecasting models. sites (SNSs). Tavana et al. (2013) had also proposed a hybrid method
Khoo and Ho (1996) integrated fuzzy set theory (Zadeh, 1965) with for the selection of social media platforms using fuzzy analytic process
QFD and proposed a framework of fuzzy QFD deployment system to and gray relations. Recently, Çalík (2020) have proposed an algorithm
design a team to select proper alternative in uncertain environment. for the selection of SNSs using best–worst method and fuzzy VIKOR.
After that many researchers (Bevilacqua et al., 2006; Lima-Junior and Proposed PFS and QFD-based MCGDM method is a general purpose
Carpinetti, 2016; Lee et al., 2015; Babbar and Amin, 2018; Dat et al., MCGDM method that analyzes users’ heterogeneous preferences in
2015) also proposed fuzzy set and QFD based methods to model un- SNSs unlike the proposals of Tavana et al. (2013) and Çalík (2020) in
certainties due to imprecise and vague information. Recently, Babar which SNSs were ranked for promoting an airline and travel agency.
and Ali (2021) and Lee and Park (2021) proposed fuzzy QFD in Rest of research paper is organized as follows. Preliminaries of PFS,
enhancement of electric vehicles’ market competitiveness and to rank basic operations on PFS along with score functions, accuracy functions
the priorities of work activities in construction for enhancing the safety. are reviewed in Section 2. This section also explains QFD process and
Atanassov (1999) defined intuitionistic fuzzy sets (IFS) as gener- presents an overview of house of quality (HOQ). Methodology of pro-
alization of fuzzy sets to include non-determinacy caused by single posed MCGDM is explained using various steps in Section 3. Proposed
function for both membership and non-membership grades. Although PFS and QFD-based MCGDM method is applied to a real-life problem
fuzzy QFD had an edge over conventional QFD in handling uncer- of selection of SNSs in Section 4 to understand its implementation.
tainty due to imprecision and linguistic representation of preferences of Section 5 includes comparison of developed method with conventional
customer or decision makers, they were found inefficient in handling QFD and IFS-based QFD method. This section also includes the vali-
non-determinacy in the information provided by decision makers. Li dation of proposed MCGDM method. Results of the study taken in the
et al. (2014) and Jian et al. (2016) integrated IFS with QFD and research paper are discussed in Section 6 and finally, conclusions are
proposed MCDM methods with non-deterministic information of ratings presented in Section 7.
of alternatives and criteria. Gitinavard et al. (2016), Wu et al. (2017),
Shi et al. (2018), Aikhuele (2017), and Haiyun et al. (2021) proposed 2. Preliminaries
various QFD based MCDM methods using interval-valued IFS, hesitant
fuzzy set and hesitant interval-valued IFS. Recently, Haktanír et al. In this section, we briefly review the definitions of PFS, basic
(2021) integrated spherical fuzzy set and QFD for designing of drone. operations on PFS, score functions, accuracy functions and an overview
Fuzzy set, IFS and their extensions are limited to model uncertainty of QFD and HOQ.
and non-determinism due to imprecise information and cannot model
the uncertainty when the information is in form of ‘‘yes’’, ‘‘no’’, ‘‘re- 2.1. Picture fuzzy set, basic operations, score, and accuracy functions
fusal’’ and ‘‘abstain’’. This specific type of information is used very
frequently in many real-life problems of decision making in data anal- Definition 1. Let X be a reference set. A PFS, A on X is defined as
ysis, survey analysis and voting system where voters may be divided mathematical object of the following form.
into aforesaid four categories. As an IFS considers both ‘‘refusal’’ and
‘‘abstain’’ equivalent, Cuong (2013a,b) extended IFS to picture fuzzy set 𝐴 = {⟨𝑥, 𝜇𝐴 (𝑥), 𝜂𝐴 (𝑥), 𝜈𝐴 (𝑥)⟩ |𝑥 ∈ 𝑋} (1)
(PFS) to model uncertainty and non-determinism due aforesaid men-
Here, 𝜇𝐴 (𝑥), 𝜂𝐴 (𝑥) and 𝜈𝐴 (𝑥) denote positive membership, neutral
tioned information. Singh (2015), Wei (2017, 2018)) proposed picture
membership, and negative membership grade of x in A respectively
fuzzy aggregation operator and picture fuzzy Hamacher aggregation
with condition 𝜇𝐴 (𝑥) + 𝜂𝐴 (𝑥) + 𝜈𝐴 (𝑥) ≤ 1∀𝑥 ∈ 𝑋.𝜌𝐴 (𝑥) = 1 − (𝜇𝐴 (𝑥) +
operator for MCGDM problems. Joshi and Kumar (2018) and Zeng
𝜂𝐴 (𝑥) + 𝜈𝐴 (𝑥)) defines degree of refusal. It is important to note that in
et al. (2019) defined similarity measure for PFS and linguistic PFS
the absence of degree of refusal, PFS converts to an IFS. For sake of
for MCGDM problems. Singh and Kumar (2020) defined picture fuzzy
convenience 𝑝 = (𝜇, 𝜂, 𝜈, 𝜌) is called picture fuzzy element (PFE), where
Choquet integral operator to model interaction phenomena among
𝜇, 𝜂, 𝜈, 𝜌 ∈ [0, 1] and 𝜇 + 𝜂 + 𝜈 + 𝜌 = 1
decision criteria in VIKOR.
Cuong (2013a,b) defined following operational laws for two PFEs
Correlations among engineering criteria (ECs) and customer criteria
𝑝1 = ⟨𝜇1 , 𝜂1 , 𝜈1 ⟩𝑝2 = ⟨𝜇2 , 𝜂2 , 𝜈2 ⟩.
(CRs) in QFD indicate the inner dependence among them. Correlation
{ }
derives overall relationship between CRs and ECs to determine the 1. 𝑝1 ⊕ 𝑝2 = 1 − (1 − 𝜇1 )(1 − 𝜇2 ), 𝜂1 𝜂2 , 𝜈1 𝜈2
importance of ECs in the assessment of alternatives and pay a pivot role { }
2. 𝑝1 ⊗ 𝑝2 = 𝜇1 𝜇2 , 1 − (1 − 𝜂1 )(1 − 𝜂2 ), 1 − (1 − 𝜈1 )(1 − 𝜈2 )
in QFD. The correlation among ECs itself can be positive and negative. { }
3. 𝜆𝑝1 = 1 − (1 − 𝜇1 )𝜆 , 𝜂1𝜆 , 𝜈1𝜆
Few ECs may also be not related to each other and no correlation can { }
4. 𝑝𝜆1 = 𝜇1𝜆 , 1 − (1 − 𝜂1 )𝜆 , 1 − (1 − 𝜈1 )𝜆
also be found among them. Positive and negative correlations reflect
support and conflict among ECs. Positive correlation is easier to manage 5. 𝑝1 ⊕ 𝑝2 = 𝑝2 ⊕ 𝑝1
than negative correlation in QFD and many researchers (Wu et al., 6. 𝜆1 (𝑝1 ⊕ 𝑝2 ) = 𝜆1 𝑝1 ⊕ 𝜆1 𝑝2 , for any 𝜆1 , 𝜆2 ≥ 0
2017; Gitinavard et al., 2016; Shi et al., 2018; Aikhuele, 2017; Li 7. (𝜆1 + 𝜆2 )𝑝1 = 𝜆1 𝑝1 ⊕ 𝜆2 𝑝1 , for any 𝜆1 , 𝜆2 ≥ 0
et al., 2014) have proposed various methodologies to deal with positive
correlation. Although normalization (Dong et al., 2017) that needs Definition 2. Let 𝑝1 , 𝑝2 , . . . . . . .., 𝑝n be n PFEs. Wei (2017) defined
maximum and minimum values of ECs is used for negative correlation. picture fuzzy weighted average (PFWA) and picture fuzzy weighted
Since in some circumstances is not possible to have maximum and geometric (PFWG) aggregation operators as follows.
minimum value for each EC, we provide a novel approach to deal ( )
∏ 𝑛 ∏
𝑛 ∏
𝑛
with positive, negative and no correlations among ECs. Motivated with 𝑃 𝐹 𝑊 𝐴𝜔 (𝑝1 , 𝑝2 , … ....𝑝𝑛 ) = (1 − (1 − 𝜇𝑖 ))𝜔𝑖 , (𝜂𝑖 )𝜔𝑖 , (𝜈𝑖 )𝜔𝑖 (2)
the advantages of PFS over fuzzy set and IFS in MCGDM methods, we 𝑖=1 𝑖=1 𝑖=1
integrate PFS with QFD to propose a novel MCGDM method. 𝑃 𝐹 𝑊 𝐺𝜔 (𝑝1 , 𝑝2 , … ....𝑝𝑛 )

2
A. Singh and S. Kumar Engineering Applications of Artificial Intelligence 104 (2021) 104395
( )

𝑛 ∏
𝑛 ∏
𝑛
= (𝜇𝑖 )𝜔𝑖 , (1 − (1 − 𝜂𝑖 ))𝜔𝑖 , (1 − (1 − 𝜈𝑖 ))𝜔𝑖 (3)
𝑖=1 𝑖=1 𝑖=1

Definition 3. Let 𝑝 = (𝜇, 𝜂, 𝜈) be a PFE. Garg (2017) defined score


function S(p) and accuracy function H (p) as follows.

𝑆(𝑝) = 𝜇 − 𝜂 − 𝜈 (4)
𝐻(𝑝) = 𝜇 + 𝜂 + 𝜈 (5)
Here, 𝑆(𝑝) ∈ [−1, 1] and 𝐻(𝑝) ∈ [0, 1]. Two PFEs, 𝑝1 and 𝑝2 can be
compared using score and accuracy function according to the following
rule.
(1) If𝑆(𝑝1 ) > 𝑆(𝑝2 ) then 𝑝1 is superior to 𝑝2 and it is denoted as 𝑝1 > 𝑝2
(2) If 𝑆(𝑝1 ) = 𝑆(𝑝2 ) then
(i) 𝐻(𝑝1 ) = 𝐻(𝑝2 ) implies that is 𝑝1 is equivalent to 𝑝2 and it is denoted
as 𝑝1 ∼ 𝑝2 .
(ii) 𝐻(𝑝1 ) > 𝐻(𝑝2 ) implies that 𝑝1 is superior to 𝑝2 and it is denoted as
𝑝1 >𝑝2 .

2.2. QFD and house of quality

QFD technique is well-known practical problem-solving method that


addresses design quality challenges to meet better customers’ need and
expectations. QFD was developed at Mitsubishi Company in Japan and
later Toyota and other company adopted its methodology to facilitate
product development process. Whole QFD process is accomplished in
Fig. 1. Structure of HOQ.
four phases of product planning, part deployment, process planning
and production planning phase which are linked as a chain and output
of one phase is employed as input for the next phase. Product plan- { } { }
ning phase of QFD was named as house of quality (HOQ) by Hauser Let 𝐴 = 𝐴1 , 𝐴2 , … ...𝐴𝑚 , 𝐶𝑅 = 𝐶𝑅1 , 𝐶𝑅2 , … ..𝐶𝑅𝑝 , 𝐸𝐶 =
{ } { }
and Clausing (1988). HOQ is used to translate qualitative customer 𝐸𝐶1 , 𝐸𝐶2 , … ...𝐸𝐶𝑞 and 𝐷 = 𝐷1 , 𝐷2 , … ...𝐷𝑡 be sets of competing
requirements into measurable engineering characteristics and prioritize alternatives, customer criteria, engineering characteristics and decision
their importance and is considered as the key element. The integral makers. Let 𝜆 = (𝜆1 , 𝜆2 ......., 𝜆𝑘 ) be the set of preference weights of
components or rooms of typical HOQ structure (Fig. 1) are explained decision makers. In the proposal of the proposed MCGDM method, we
briefly as follows. suppose the following.
(i) User/Customer requirements (CRs): Customers’ needs are ( ) ( )
(𝑘) (𝑘) (𝑘)
I. 𝑅̂ 𝑘 = 𝑟̂(𝑘) 𝑝×𝑞 = 𝜇𝑟,𝑖𝑗 , 𝜂𝑟,𝑖𝑗 , 𝜈𝑟,𝑖𝑗 is linguistic decision matrix
identified to construct HOQ to understand their expectations. 𝑝×𝑞
(ii) System/Engineering characteristics (ECs): Technical of relationship between
( CR𝑖 and) EC 𝑗 .
( ) (𝑘) (𝑘) (𝑘)
attributes of product and substitute quality characteristic of II. 𝐶̂ 𝑘 = 𝑐̂(𝑘) 𝑞×𝑞 = 𝜇𝑐,𝑖𝑗 , 𝜂𝑐,𝑖𝑗 , 𝜈𝑐,𝑖𝑗 is linguistic decision matrix
𝑞×𝑞
customers’ requirements are identified as means to fulfill CRs. of correlation
( ) between(EC 𝑖 and EC 𝑗 . )
(iii) Importance of CRs: As all CRs may not be equally important for III. 𝐴̂ 𝑘 = 𝑎̂(𝑘) (𝑘)
= 𝜇𝑎,𝑖𝑗 (𝑘)
, 𝜂𝑎,𝑖𝑗 (𝑘)
, 𝜈𝑎,𝑖𝑗 is linguistic decision
𝑖𝑗
customers, they are prioritized by direct evaluation or by using 𝑚×𝑝 𝑚×𝑝
matrix of the rating of alternative 𝐴𝑖 with respect to customer
different analytical techniques.
criterion CR𝑗 .
(iv) Relationship matrix between CRs and ECs: Relationship ma-
trix between CRs and ECs represents intensity of relation be- IV. 𝑊̂ 𝐾 = (𝑤̂ (𝑘) (𝑘) (𝑘) (𝑘)
𝑗 )1×𝑝 = (𝜇𝑤,𝑗 , 𝜂𝑤,𝑗 , 𝜈𝑤,𝑗 )1×𝑝 is linguistic decision matrix
tween them and identifies the extent by which ECs affects as- of the weight of customer criterion CR𝑗 provided by decision
sociated CRs. maker 𝐷𝑘 .
(v) Correlation between ECs: The correlation matrix shows the in-
terdependence among ECs and helps in quantifying the possible Proposed QFD and PFS based MCGDM method includes following
interaction among ECs for future trade off decisions. steps.
(vi) Assessment of alternatives in terms of CRs: This matrix eval- Step 1: Transform linguistic information provided by decision makers
uates competing alternatives with respect to CRs by direct cus- into picture fuzzy numbers.
tomer survey.
Step 2: Aggregate assessment of each decision maker with respect to
(vii) Importance of ECs: Importance of ECs is determined using
̂ of each alternative with respect
CRs to compute collective ratings (𝐴)
assessment of alternatives in terms of CRs. Ranking of ECs il-
lustrates which EC has greater effect on customer’s satisfaction. to CRs. Also, aggregate opinion of decision makers about the impor-
tance of CRs to compute and collective weights (𝑊̂ ) using following
aggregation operators.
3. Proposed QFD and PFS based MCGDM method
𝐴̂ = (𝜇𝑎,𝑖𝑗 , 𝜂𝑎,𝑖𝑗 , 𝜈𝑎,𝑖𝑗 )
( 𝑡 )
QFD is well recognized and comprehensive method which is widely ∏ ( (𝑘) )𝜆𝑘 𝑡 (
∏ (𝑘)
) 𝜆𝑘 𝑡 (
∏ (𝑘)
) 𝜆𝑘
used to meet customers’ expectation by translating customer require- = 𝜇𝑎,𝑖𝑗 ,1 − 1 − 𝜂𝑎,𝑖𝑗 ,1 − 1 − 𝜈𝑎,𝑖𝑗 (6)
ments into design requirements. PFS is more generalized and effec- ( )
𝑘=1 𝑘=1 𝑘=1
tive tool than fuzzy set and IFS in handling uncertainty and non- 𝑊̂ = 𝑤̂ 𝑗 = (𝜇𝑤,𝑗 , 𝜂𝑤,𝑗 , 𝜈𝑤,𝑗 )
( 𝑡 )
determinacy in acquired information. Integrating inherent characteris- ∏ ( (𝑘) )𝜆𝑘 ∏𝑡 (
(𝑘)
) 𝜆𝑘 𝑡 (
∏ (𝑘)
) 𝜆𝑘
tics of both QFD, we propose a MCGDM method and PFS in this section = 𝜇𝑤,𝑗 ,1 − 1 − 𝜂𝑤,𝑗 ,1 − 1 − 𝜈𝑤,𝑗 (7)
follows. 𝑘=1 𝑘=1 𝑘=1

3
A. Singh and S. Kumar Engineering Applications of Artificial Intelligence 104 (2021) 104395

Step 3: Aggregate opinions of decision makers about the relationship Table 1


∧ Linguistic terms and corresponding PFN.
between CRs and ECs to compute collective relationship 𝑅 = (𝑟𝑖𝑗 ) =
Linguistic Terms Picture fuzzy number
(𝜇𝑟, 𝑖𝑗 , 𝜂𝑟, 𝑖𝑗 , 𝜐𝑟, 𝑖𝑗 ) between them using Eq. (6).
Very Poor (VP) < 0.10, 0.00, 0.85 >
Step 4: Calculate collective correlation among ECs using following Poor (P) < 0.25, 0.05, 0.60 >
expression. Moderate poor (MP) < 0.30, 0.00, 0.60 >
( ) Fair (F) < 0.50, 0.10, 0.40 >
𝐶̂ = 𝑐̂𝑖𝑗 = (𝜇𝑐,𝑖𝑗 , 𝜂𝑐,𝑖𝑗 , 𝜈𝑐,𝑖𝑗 ) Moderate good (MG) < 0.60, 0.00, 0.30 >
( 𝑘 )
( ) 𝑘 ( ) 𝑘 ( )
1 ∑ (𝑘) 1 ∑ (𝑘) 1 ∑ (𝑘)
Good (G) < 0.75, 0.05, 0.10 >
= 𝜇𝑐,𝑖𝑗 , 𝜂𝑐,𝑖𝑗 , 𝜈𝑐,𝑖𝑗 (8) Very good (VG) < 0.90, 0.00, 0.05 >
𝑘 𝑘=1 𝑘 𝑘=1 𝑘 𝑘=1

Step 5: Calculate overall impact on each EC due to correlations with Table 2


other ECs to determine magnitude of their interdependency (support or Linguistic terms for correlations and corresponding PFN.
contradict) using following expression. Nature of Correlation Linguistic term Picture fuzzy number
( ) Very good (VG− ) < −0.90, 0.00, −0.10 >
𝑂̂ = 𝑜̂𝑗 = (𝜇𝑜,𝑖 , 𝜂𝑜,𝑖 , 𝜈𝑜,𝑖 )
Good (𝐺− ) < −0.75, −0.05, −0.10 >
⎛ ⎞ Moderate Good (MG− ) < −0.60, 0.00, −0.30 >
⎜ 1 ∑( 1 ∑( 1 ∑(
𝑞 𝑞 𝑞
) ) )⎟ Negative Correlation
Fair (𝐹− ) < −0.50, −0.10, −0.45 >
=⎜ 𝜇𝑐,𝑖𝑗 , 𝜂𝑐,𝑖𝑗 , 𝜈𝑐,𝑖𝑗 ⎟ (9) Moderate Poor (MP− ) <-0.30, 0.00, −0.55 >
⎜ 𝑞 − 1 𝑗=1 𝑞 − 1 𝑗=1 𝑞 − 1 𝑗=1 ⎟
⎝ 𝑖≠𝑗 𝑖≠𝑗 𝑖≠𝑗 ⎠ Poor (𝑃− ) < −0.25, −0.05, −0.75 >
Very Poor (VP− ) < −0.10, 0.00, −0.85 >
Step 6: Since Correlation of ECs affects the relationship between CRs
No Correlation < 0.00, 0.00, 0.00 >
and ECs, therefore overall relationship between CRs and ECs needs to
Very Poor (VP+ ) < 0.10,0.00,0.85 >
be determined using both relationship between CRs and ECs and the
Poor (𝑃+ ) < 0.25,0.05,0.60 >
correlation among ECs. Compute overall relationship between CRs and Moderate Poor (MP+ ) < 0.30,0.00,0.60>
Positive correlation
ECs using following expression. Fair (𝐹+ ) < 0.50,0.10,0.40>
( ) Moderate Good (MG+ ) < 0.60,0.00,0.30>
𝑅̂ ′ = 𝑟̂′𝑖𝑗 = (𝜇𝑟′′ ,𝑖𝑗 , 𝜂𝑟′′ ,𝑖𝑗 , 𝜈𝑟′′ ,𝑖𝑗 ) Good (𝐺+ ) < 0.75,0.05,0.10>
(( ) ( ) ( )) Very Good (VG+ ) < 0.90,0.00,0.05>
𝜇𝑟,𝑖𝑗 + 𝜇𝑜,𝑗 𝜂𝑟,𝑖𝑗 + 𝜂𝑜,𝑗 𝜈𝑟,𝑖𝑗 + 𝜈𝑜,𝑗
= , , (10)
2 2 2

Step 7: Since weights of CRs given by the customers are transformed translates the importance of customer expectations into engineering
into weights of ECs using overall relationships matrix, weights of ECs characteristics to reduce the computational cost. In this section, we
are computed using following expression. implement proposed MCGDM method in evaluating performance of
( )
Facebook (𝐴1 ), Instagram (𝐴2 ), WhatsApp (𝐴3 ) and Twitter (𝐴4 ) by
𝑊̂ ′ = 𝑤̂ ′𝑗 = (𝜇𝑤′ ,𝑗 , 𝜂𝑤′ ,𝑗 , 𝜈𝑤′ ,𝑗 )
(( ) ( ) ( )) group of eight users/decision makers {𝐷1 , 𝐷2 , 𝐷3 , 𝐷4 , 𝐷5 , 𝐷6 , 𝐷7 , 𝐷8 }
∑𝑝
𝜇𝑟′ ,𝑖𝑗 + 𝜇𝑤,𝑖 − 𝜇𝑟′ ,𝑖𝑗 𝜇𝑤,𝑖 𝜂𝑟′ ,𝑖𝑗 𝜂𝑤,𝑖 𝜈𝑟′ ,𝑖𝑗 𝜈𝑤,𝑖 using ten criteria which are divided into following CRs and ECs.
= , , (11)
𝑖=1
𝑝 𝑝 𝑝
(1) Customer requirements: From the analysis of the questionnaire
Use score function 𝑆𝑤′ ,𝑖 = 𝜇𝑤′ ,𝑖 − 𝜂𝑤′ ,𝑖 − 𝜈𝑤′ ,𝑖 to prioritize ECs that and interview following customer criteria are selected.
determines which EC has more influence on CR.
Step 8: Transform ratings of alternatives with respect to CRs into rat- (i) Easy to use (𝐶𝑅1 )
ings with
( respect
) to ECs using overall(relationship
) matrix and calculate (ii) Multitasking (𝐶𝑅2 )
∧ ∧
ratings 𝐴′ and weighted ratings 𝐹 of alternatives on ECs using (iii) Storage friendly (𝐶𝑅3 )
(iv) Easy in developing network (𝐶𝑅4 )
following expressions.
(v) Security and privacy (𝐶𝑅5 )
𝐴̂ ′ = (𝜇𝑎′ ,𝑖𝑗 , 𝜂𝑎′ ,𝑖𝑗 , 𝜈𝑎′ ,𝑖𝑗 )
(( ) ( ) ( )) (2) Engineering characteristics (System criteria): In the present
∑𝑝
𝜇𝑟′ ,𝑙𝑗 + 𝜇𝑎,𝑖𝑙 − 𝜇𝑟′ ,𝑙𝑗 𝜇𝑎,𝑖𝑙 𝜂𝑟′ ,𝑙𝑗 𝜂𝑎,𝑖𝑙 𝜈𝑟′ ,𝑙𝑗 𝜈𝑎,𝑖𝑙
= , , (12) experimental study, following technical criteria which are called
𝑙=1
𝑝 𝑝 𝑝 ECs are selected.
( )
𝐹̂ = (𝜇𝑓 ,𝑖𝑗 , 𝜂,𝑖𝑗 , 𝜈𝑓 ,𝑖𝑗 ) = 𝜇𝑎′ ,𝑖𝑗 + 𝜇𝑤′ ,𝑗 − 𝜇𝑎′ ,𝑖𝑗 𝜇𝑤′ ,𝑗 , 𝜂𝑎′ ,𝑖𝑗 𝜂𝑤′ ,𝑗 , 𝜈𝑎′ ,𝑖𝑗 𝜈𝑤′ ,𝑗
(i) Structure of sites (𝐸𝐶 1 )
(13) (ii) Number of associated features (𝐸𝐶 2 )
Step 9: Apply PFWG (Eq. (3)) on weighted picture fuzzy number and (iii) Way of accessing site and storing data (𝐸𝐶 3 )
rank the alternatives using score function value (Eq. (4)) in accordance (iv) Policy of making network (𝐸𝐶 4 )
in decreasing order. (v) Use of updated security technology (𝐸𝐶 5 )

4. Implementation of proposed MCGDM method in selection of {𝐷1 , 𝐷2 , 𝐷3 , 𝐷4 } and {𝐷5 , 𝐷6 , 𝐷7 , 𝐷8 } symbolize sets of four users
social networking site and four analysts who are taken as decision makers in this study. Users
𝐷𝑖 (𝑖 = 1 … 4) provide their opinions about the weight of CRs. They also
Social network is the Internet-based social media sites which are provide their linguistic ratings to evaluate a SNS with respect to CRs
used to stay connected with friends, family, colleagues, customers, or which are converted into PFNs using Table 1. Analysts 𝐷𝑖 (𝑖 = 5 … 8)
clients for social purpose or business purpose, or both. During last provide linguistic rating to the relationship between CRs and ECs which
decades and till now, social network has expanded with amazing rate are converted into PFNs using Table 2. All above mentioned linguistic
because of its innovative uses in finding new jobs, getting feedback ratings and relationship matrices are shown in Tables 3–6.
on ideas and products, spreading awareness, and promoting new re- Following are stepwise computations for ranking SNSs using pro-
search and as marketing tools to increase sales. Selection of a SNS posed PFS and QFD approach based MCGDM method.
depends upon the numbers of interdependent criteria. QFD divides Step 1: Linguistic information provided by decision makers are trans-
them into groups of customer criteria and engineering criteria and formed into PFNs using Tables 1 and 2.

4
A. Singh and S. Kumar Engineering Applications of Artificial Intelligence 104 (2021) 104395

Table 3 Preference order of ECs in accordance of having greater influence on


Linguistic rating of weights of CRs.
the customers’ satisfaction is as obtained as 𝐸𝐶 2 > 𝐸𝐶 3 > 𝐸𝐶 4 > 𝐸𝐶 5 >
𝐶𝑅1 𝐶𝑅2 𝐶𝑅3 𝐶𝑅4 𝐶𝑅5
𝐸𝐶 1 . It shows that the most and the least attention needs to be paid on
𝐷1 G VG G G VG
number of associated features (𝐸𝐶 2 ) and on way of accessing site and
𝐷2 G MP F MG G
𝐷3 MG F G MP VG storing data (𝐸𝐶 3 ) in order to satisfy the customers’ requirement.
𝐷4 G MP G G VG Step 8: Ratings and weighted ratings of alternatives with respect to ECs
(Tables 12 & 13) are computed using Eq. (12) and Eq. (13).
Step 9: Performance of SNSs with respect to ECs is aggregated using
Step 2: Preferences of each user are aggregated using Eq. (6) to PFWG aggregation operator (Eq. (3)) and corresponding score func-
determine collective ratings of each SNS with respect to CRs (Table 7). tion value is computed using Eq. (4). Table 14 shows aggregated
During aggregation all preferences of each user is considered equally PFNs expressing performance of each SNS with respect to ECs with
important by considering 𝜆𝑘 = 1∕4(𝑘 = 1, 2, 3, 4). corresponding score function values.
Collective weights of CRs are computed using Eq. (7) and is shown In accordance of decreasing score function values, the SNSs are
as following weight matrix. ranked as follows.

⎛𝑤1 ⎞ ⎛0.709, 0.038, 0.155⎞ Facebook (𝐴1 ) > Instagram (𝐴2 ) > WhatsApp (𝐴3 ) > Twitter (𝐴4 )
⎜∧ ⎟ ⎜ ⎟
⎜𝑤2 ⎟ ⎜0.449, 0.026, 0.450⎟
∧ ⎜∧ ⎟ ⎜ ⎟
𝑊 = ⎜𝑤3 ⎟ = ⎜0.678, 0.063, 0.187⎟ 5. Comparative analysis and validation of proposed MCGDM
⎜ ⎟ ⎜ ⎟ method
⎜∧ ⎟ ⎜ ⎟
⎜𝑤4 ⎟ ⎜0.564, 0.025, 0.310⎟
⎜∧⎟ ⎜ ⎟
⎝𝑤5 ⎠ ⎝0.860, 0.013, 0.063⎠ To test the capability, advantages of integration of QFD with PFS
Step 3: Relationship between CRs and ECs are aggregated using Eq. (6) and influence of parametric uncertainty, we determine ranking of SNSs
to compute collective relationship (Table 8) between them. During using conventional QFD and IFS-based QFD approach to compare the
aggregation preferences of each analyst is considered equally important ranking results. Crisp numerical values (Table 15) are assigned to
by considering 𝜆𝑘 = 1∕4(𝑘 = 1, 2, 3, 4). linguistic term in implementing conventional QFD.
Step 4: Collective correlations among ECs (Table 9) are calculated using Assessment of the SNSs by the different users are aggregated using
Eq. (8). average operator to have aggregated matrices. After performing all
Step 5: Overall impact on each EC due to correlations with other ECs steps of QFD, the preference order of ECs is obtained as 𝐸𝐶 3 > 𝐸𝐶 2 >
is calculated using Eq. (9) and is shown by following matrix. 𝐸𝐶 5 > 𝐸𝐶 4 > 𝐸𝐶 1 and eventually SNSs are ranked as 𝐴1 > 𝐴3 > 𝐴2 >
𝐴4 .
⎛𝐸𝐶1 ⎞ ⎛ −0.016 −0.0094 0.2406 ⎞
⎜ ⎟ ⎜ ⎟ Further, IFS-based QFD is also applied for the ranking of SNSs by
⎜𝐸𝐶2 ⎟ ⎜ −0.084 0.0063 0.0875 ⎟ assigning intuitionistic fuzzy numbers to linguistic terms (Table 16).
⎜ ⎟ ⎜ ⎟
⎜𝐸𝐶 ⎟ = ⎜ 0.3781 0.0344 0.5250 ⎟ Assessment of the SNSs by the different users are aggregated using
⎜ 3⎟ ⎜ ⎟ following aggregation operator (Xu, 2007).
⎜ ⎟ ⎜ ⎟
⎜𝐸𝐶4 ⎟ ⎜ 0.1500 0.0063 0.3094 ⎟ ( 𝑡 )
⎜ ⎟ ⎜ ⎟ ∏ ( (𝑘) )𝜆𝑘 𝑡 (
∏ (𝑘)
)𝜆𝑘
⎝𝐸𝐶5 ⎠ ⎝−0.1310 −0.0060 −0.0690⎠ 𝐴̂ = (𝜇𝑎,𝑖𝑗 , 𝜈𝑎,𝑖𝑗 ) = 𝜇𝑎,𝑖𝑗 ,1 − 1 − 𝜈𝑎,𝑖𝑗
𝑘=1 𝑘=1
Step 6: Overall relationship between CRs and ECs (Table 10) is deter-
mined using Eq. (10). Here, 𝜆𝑘 is preference weight of 𝑘th decision maker. After perform-
Step 7: Since preferences given by users are transformed into pref- ing all steps of QFD, the preference order of ECs is obtained as 𝐸𝐶 5 >
erences of analysts, weights of ECs (Table 11) are calculated using 𝐸𝐶 1 > 𝐸𝐶 3 > 𝐸𝐶 4 > 𝐸𝐶 2 to give ranking 𝐴1 > 𝐴2 > 𝐴3 > 𝐴4 of SNSs.
overall relationship matrix and Eq. (11). Using score function ECs are Preferences of ECs and ranking of the SNSs which are obtained using
prioritized. Table 11 shows weights of the ECs along with their priority conventional QFD, intuitionistic fuzzy QFD and proposed PFS-based
order. QFD are exhibited in Table 16.

Table 4
Linguistic rating of the alternatives with respect to CRs.
𝐷1 𝐷2 𝐷3 𝐷4
𝐴1 𝐴2 𝐴3 𝐴4 𝐴1 𝐴2 𝐴3 𝐴4 𝐴1 𝐴2 𝐴3 𝐴4 𝐴1 𝐴2 𝐴3 𝐴4
𝐶𝑅1 VG G VG MG MP MG VG P F MG VG F G MG VG F
𝐶𝑅2 G MP VP P G MG MP VP MG G F MP G P G MP
𝐶𝑅3 G G MP MG MG G G MP VG VG MP P MG F MG P
𝐶𝑅4 MG F P F VG G MG G VG MG P MG VG F MG G
𝐶𝑅5 F F G MG MP MG G F VP G VG MG MG MG G F

Table 5
Linguistic rating of relationship between the CRs and ECs.
𝐷5 𝐷6 𝐷7 𝐷8
𝐶𝑅1 𝐶𝑅2 𝐶𝑅3 𝐶𝑅4 𝐶𝑅5 𝐶𝑅1 𝐶𝑅2 𝐶𝑅3 𝐶𝑅4 𝐶𝑅5 𝐶𝑅1 𝐶𝑅2 𝐶𝑅3 𝐶𝑅4 𝐶𝑅5 𝐶𝑅1 𝐶𝑅2 𝐶𝑅3 𝐶𝑅4 𝐶𝑅5
𝐸𝐶 1 MG G MG MG G G MP P P P VG VP VP VP VP VG P P P P
𝐸𝐶 2 G VG MG F G MP MP MP MG F P VG VP MG F F G MG MP MP
𝐸𝐶 3 G G MP F VG MG F MG G MG F VP VG VP VP G MP G MP F
𝐸𝐶 4 F G MP G VG MP MP F MG MG P VP VP VG G F P P G P
𝐸𝐶 5 G G MG MP VG P P P MG G VP VP P P VG F F MG G G

5
A. Singh and S. Kumar Engineering Applications of Artificial Intelligence 104 (2021) 104395

Fig. 2. HOQ structure for conventional QFD for case taken in the study.

5.1. Validity of proposed MCGDM method Test Criterion 1. An effective MCGDM method does not change the
indication of the best alternative on replacing a non-optimal alternative
by another worse alternative.
However, validating a MCGDM method has always been challenging Test Criterion 2. When a decision-making problem is decomposed into
due to self-inconsistency or cognitive bias during preferences elicitation smaller problems and same MCGDM method is applied on smaller
process, Wang and Triantaphyllou (2008) established following test problems to rank the alternatives, combined ranking of the alternatives
criteria. should be same as original un-decomposed decision-making problem.

Table 6
Linguistic rating of correlation among ECs.
𝐷5 𝐷6 𝐷7
𝐸𝐶 1 𝐸𝐶 2 𝐸𝐶 3 𝐸𝐶 4 𝐸𝐶 5 𝐸𝐶 1 𝐸𝐶 2 𝐸𝐶 3 𝐸𝐶 4 𝐸𝐶 5 𝐸𝐶 1 𝐸𝐶 2 𝐸𝐶 3 𝐸𝐶 4 𝐸𝐶 5
𝐸𝐶 1 VG+ 𝐺− 𝐺+ 𝐹+ – VG+ MG− MP+ 𝑃+ – VG+ VG− VP+ VP+ –
𝐸𝐶 2 𝐺− VG+ 𝐹+ MG+ 𝐺− MG− VG+ 𝐹+ VP+ 𝐹− VG− VG+ VP+ MG+ VP−
𝐸𝐶 3 𝐺+ 𝐹+ VG+ MG+ VL+ MP+ 𝐹+ VG+ 𝑃+ 𝐹+ VP+ VP+ VG+ VP+ VP+
𝐸𝐶 4 𝐹+ MG+ MG+ VG+ 𝑃− 𝑃+ VP+ 𝑃+ VG+ MP− VP+ MG+ VP+ VG+ MG−
𝐸𝐶 5 – 𝐺− VL+ 𝑉 𝑃+ VG+ – 𝐹− 𝐹+ MP− VG+ – VP− VP+ MG− VG+

𝐷8
𝐸𝐶 1 𝐸𝐶 2 𝐸𝐶 3 𝐸𝐶 4 𝐸𝐶 5
VGP MG− MP+ MP+ –
MG− VGP 𝐹+ MP+ MP−
MP+ 𝐹+ VGP MG+ MG+
MP+ MP+ MG+ VGP 𝐺−
– MP− MG+ 𝐺− VGP

Table 7
Collective ratings of the alternative with respect to CRs.
𝐶𝑅1 𝐶𝑅2 𝐶𝑅3 𝐶𝑅4 𝐶𝑅5
𝐴1 (0.564, 0.038, 0.327) (0.709, 0.038, 0.155 (0.702, 0.013, 0.195) (0.813, 0, 0.12) (0.308, 0.026, 0.602)
𝐴2 (0.634, 0.013, 0.256) (0.429, 0.025, 0.437) (0.709, 0.051, 0.176) (0.579, 0.063, 0.31) (0.606, 0.038, 0.283)
𝐴3 (0.9, 0, 0.05) (0.326, 0.038, 0.576) (0.449, 0.013, 0.437) (0.381, 0.025, 0.471) (0.785, 0.038, 0.283)
𝐴4 (0.44, 0.063, 0.437) (0.228, 0, 0.687) (0.326, 0.028, 0.54) (0.641, 0.051, 0.236) (0.548, 0.051, 0.352)

6
A. Singh and S. Kumar Engineering Applications of Artificial Intelligence 104 (2021) 104395

Table 8
Collective relationships between CRs and ECs.
𝐸𝐶 1 𝐸𝐶 2 𝐸𝐶 3 𝐸𝐶 4 𝐸𝐶 5
𝐶𝑅1 (0.777, 0.013, 0.132) (0.41, 0.051, 0.458) (0.641, 0.051, 0.236) (0.37, 0.063, 0.51) (0.311, 0.051, 0.576)
𝐶𝑅2 (0.274, 0.025, 0.617) (0.653, 0.013, 0.245) (0.326, 0.038, 0.576) (0.274, 0.025, 0.617) (0.311, 0.051, 0.576)
𝐶𝑅3 (0.247, 0.025, 0.64) (0.322, 0, 0.586) (0.671, 0.038, 0.226) (0.247, 0.038, 0.617) (0.387, 0.025, 0.471)
𝐶𝑅4 (0.247, 0.025, 0.64) (0.482, 0.026, 0.414) (0.287, 0.013, 0.617) (0.742, 0.025, 0.301) (0.429, 0.025, 0.437)
𝐶𝑅5 (0.262, 0.038, 0.617) (0.487, 0.063, 0.4) (0.405, 0.026, 0.505) (0.564, 0.025, 0.301) (0.822, 0.025, 0.088)

Table 9
Collective correlations among ECs.
𝐸𝐶 1 𝐸𝐶 2 𝐸𝐶 3 𝐸𝐶 4 𝐸𝐶 5
𝐸𝐶 1 (0.9, 0, 0.05) (−0.713, −0.0125, −0.188) (0.363, 0.0125, 0.538) (0.288, 0.0375, 0.613) –
𝐸𝐶 2 (−0.713, −0.0125, −0.188) (0.9, 0, 0.05) (0.4, 0.075, 0.5123) (0.4, 0, 0.513) (−0.413, −0.0375, −0.488)
𝐸𝐶 3 (0.363, 0.0125, 0.538) (0.4, 0.075, 0.5123) (0.9, 0, 0.05) (0.388, 0.0125, 0.513) (0.363, 0.0375, 0.613)
𝐸𝐶 4 (0.288, 0.0375, 0.613) (0.4, 0, 0.513) (0.388, 0.0125, 0.513) (0.9, 0, 0.05) (−0.475, −0.025, −0.4)
𝐸𝐶 5 – (−0.413, −0.0375, −0.488) (0.363, 0.0375, 0.613) (−0.475, −0.025, −0.4) (0.9, 0, 0.05)

Table 10
Overall relationships between the CRs and ECs.
𝐸𝐶 1 𝐸𝐶 2 𝐸𝐶 3 𝐸𝐶 4 𝐸𝐶 5
𝐶𝑅1 (0.381, 0.011, 0.226) (0.163, 0.029, 0.287) (0.51, 0.043, 0.468) (0.26, 0.035, 0.461) (0.09, 0.022, 0.242)
𝐶𝑅2 (0.129, 0.017, 0.469) (0.284, 0.01, 0.181) (0.352, 0.036, 0.638) (0.212, 0.016, 0.515) (0.09, 0.022, 0.48)
𝐶𝑅3 (0.0116, 0.017, 0.48) (0.119, 0.003, 0.351) (0.524, 0.036, 0.463) (0.199, 0.022, 0.515) (0.128, 0.009, 0.19)
𝐶𝑅4 (0.116, 0.017, 0.48) (0.199, 0.016, 0.265) (0.333, 0.024, 0.659) (0.446, 0.016, 0.357) (0.149, 0.009, 0.173)
𝐶𝑅5 (0.123, 0.024, 0.469) (0.201, 0.035, 0.255) (0.392, 0.03, 0.603) (0.357, 0.016, 0.357) (0.345, 0.009, 0)

Table 11
Weights and priority order of ECs.
𝐸𝐶 1 𝐸𝐶 2 𝐸𝐶 3 𝐸𝐶 4 𝐸𝐶 5
Weights (0.709, 0.005, 0.103) (0.723, 0.0004, 0.058) (0.794, 0.001, 0.138) (0.752, 0.0008, 0.107) (0.698, 0.0005, 0.069)
Score values 0.606 0.665 0.655 0.645 0.629

Table 12
Ratings of the alternatives with respect to ECs.
𝐸𝐶 1 𝐸𝐶 2 𝐸𝐶 3 𝐸𝐶 4 𝐸𝐶 5
𝐴1 (0.688, 0.0004, 0.116) (0.697, 0.0005, 0.0752) (0.782, 0.0008, 0.157) (0.732, 0.0005, 0.118) (0.693, 0.0004, 0.042)
𝐴2 (0.66, 0.0007, 0.126) (0.68, 0.0006, 0.0736) (0.758, 0.0012, 0.171) (0.712, 0.0007, 0.129) (0.656, 0.0004, 0.072)
𝐴3 (0.627, 00.004, 0.152) (0.666, 0.0004, 0.0838) (0.742, 0.0007, 0.191) (0.695, 0.0004, 0.149) (0.63, 0.0003, 0.091)
𝐴4 (0.534, 0.0006, 0.192) (0.558, 0.0009, 0.1183) (0.678, 0.0013, 0.252) (0.589, 0, 0.209) (0.52, 0.0005, 0.116)

Table 13
Weighted ratings of the alternatives with respect to ECs.
𝐸𝐶 1 𝐸𝐶 2 𝐸𝐶 3 𝐸𝐶 4 𝐸𝐶 5
𝐴1 (0.909, 0, 0.012) (0.916, 0, 0.004) (0.955, 0, 0.02) (0.934, 0, 0.013) (0.907, 0, 0.003)
𝐴2 (0.901, 0, 0.013) (0.911, 0, 0.005) (0.950, 0, 0.02) (0.929, 0, 0.014) (0.896, 0, 0.005)
𝐴3 (0.891, 0, 0.016) (0.904, 0, 0.005) (0.947, 0, 0.02) (0.924, 0, 0.016) (0.888, 0, 0.006)
𝐴4 (0.864, 0, 0.020) (0.0877, 0, 0.007) (0.933, 0, 0.02) (0.898, 0, 0.022) (0.854, 0, 0.008)

Table 14 Table 15
Aggregated performance of alternatives with their score function value. Linguistic terms with corresponding numerical value and IFN.
SNN Aggregated performance with respect to EC Score function value S(𝐴i ) Linguistic Terms Crisp value Intuitionistic Fuzzy Number
𝐴1 (0.674, 0, 0.049) 0.625 Very Poor (VP) 1 < 0.10, 0.9 >
𝐴2 (0.649, 0, 0.51) 0.599 Poor (P) 2 < 0.25, 0.75 >
𝐴3 (0.629, 0, 0.56) 0.573 Moderate poor (MP) 3 < 0.4, 0.55 >
𝐴4 (0.542, 0, 0.068) 0.475 Fair (F) 4 < 0.50, 0.45 >
Moderate good (MG) 5 < 0.60, 0.30 >
Good (G) 6 < 0.75, 0.10 >
Very good (VG) 7 < 0.90, 0.10 >
To test the validity of proposed MCGDM method under test criterion
1, linguistic rating of SNSs with respects to CRs (Table 4) is modified
(Table 15) by taking complement of linguistic values of non-optimal
social networking site (𝐴2 ) (see Table 17) Further, to test the validity of proposed MCGDM method under test
After transformation of the data into PFNs and using methodology criterion 2, the original decision-making problem is decomposed into
of proposed MCGDM method, ranking of the SNSs is obtained as 𝐴1 > sets 𝐴1 , 𝐴2 , 𝐴3 and 𝐴2 , 𝐴3 , 𝐴4 of smaller decision-making problems
𝐴3 > 𝐴4 > 𝐴2 . Since Facebook (𝐴1 ) remains the best SNS with modified with original decision criteria. On applying the methodology of pro-
linguistic data of non-optimal alternative therefore, proposed MCGDM posed MCGDM method, ranking of the SNSs is obtained as 𝐴1 > 𝐴2 >
method is valid under test criterion 1. 𝐴3 and 𝐴2 > 𝐴3 > 𝐴4 . On combining these two rankings, we get the 𝐴1 >

7
A. Singh and S. Kumar Engineering Applications of Artificial Intelligence 104 (2021) 104395

Table 16 users in form of intuitionistic fuzzy number, Facebook is ranked first


Comparison of proposed PFS-based QFD.
followed by Instagram, WhatsApp, and Twitter. This may be due to
Method Preference order of ECs Ranking of SNSs
the consideration of all types of correlation among the ECs. If negative
Conventional QFD 𝐸𝐶 2 > 𝐸𝐶 3 > 𝐸𝐶 4 > 𝐸𝐶 5 > 𝐸𝐶 1 𝐴1 > 𝐴3 > 𝐴2 > 𝐴4 correlation among the ECs is discarded or handled using normalization,
QFD with IFS 𝐸𝐶 3 > 𝐸𝐶 2 > 𝐸𝐶 5 > 𝐸𝐶 4 > 𝐸𝐶 1 𝐴1 > 𝐴2 > 𝐴3 > 𝐴4
ranking of the SNSs may change.
Proposed PFS based QFD 𝐸𝐶 5 > 𝐸𝐶 1 > 𝐸𝐶 3 > 𝐸𝐶 4 > 𝐸𝐶 2 𝐴1 > 𝐴2 > 𝐴3 > 𝐴4

7. Conclusions
𝐴2 > 𝐴3 > 𝐴4 which is like the ranking of alternatives in un-decomposed
decision-making problem. Hence, proposed MCGDM method is also Being accurate and precise in decision-making with limited knowl-
valid under test criterion 2. edge and uncertain information always has been a vital issue for
researchers. In this paper, we have proposed a PFS and QFD-based
6. Results and discussion MCGDM method. The proposed MCGDM methods uses PFS to measure
the linguistic terms for qualitative analysis and characterizes fuzziness
In this section, the study carried out using feedback of the users and uncertainties more comprehensively than fuzzy set and IFS. QFD
about Facebook (𝐴1 ), Instagram (𝐴2 ), WhatsApp (𝐴3 ) and Twitter approach in proposed MCGDM method integrates heterogeneous pref-
(𝐴4 ) against CRs and ECs is presented to confirm outperformance and erences of decision makers by splitting them into the preferences of user
novelty of proposed PFS and QFD-based MCGDM method. First, we (CR) and preferences of analysts (EC). We have considered positive,
have collected information from the users regarding their requirements negative and no correlations among ECs without normalization to find
in SNSs and extent by which SNSs taken in the study meet their ex- overall effect on each EC due to correlation with other ECs.
pectations. Then, we have considered few design parameters (ECs) for Selection of a social media platform has always been an effortful
meeting the expectation of users’ requirements. We have constructed task as it needs a rigorous analysis of various interdependent criteria.
relationship matrix (Table 8) to understand impact of ECs on CRs. Since, ranking of the SNSs needs to be determined very often now-a-
For example, 𝐶𝑅1 (easy to use) feature of a SNS will also affect days, we have applied proposed MCGDM methods to rank Facebook,
𝐸𝐶 1 (structure of SNS) strongly while features 𝐸𝐶 4 (Policy of making WhatsApp, Instagram, and Twitter using ten criteria and a group of
network) and 𝐸𝐶 5 (Use of updated security technology) of a SNS also eight decision makers. Feedback from the users about these four SNSs
affect the overall operating of SNS but not very strongly. against criteria is taken in linguistic form and is converted into PFNs
In the proposed PFS and QFD-based MCGDM method, we also to model fuzziness and uncertainty more effectively. Proposed MCGDM
interpretate interdependency among ECs. For example, strong negative method ranks Facebook as the most preferable SNS based on feedback
relation between 𝐸𝐶 1 (Structure of SNS) and 𝐸𝐶 2 (Number of associ- from the users. Although Instagram was found to have an edge over
ated features in SNS) is observed. These two ECs do not support each WhatsApp, performance of both Instagram and WhatsApp against the
other as enhancement in features associated with a SNS will make its criteria taken in the study was found more over same using proposed
structure complex. Similarly, positive relation between 𝐸𝐶 4 (Policy of MCGDM method.
making network) and 𝐸𝐶 1 (Structure of site) is found, Although, it is
moderately poor but 𝐸𝐶 4 and 𝐸𝐶 1 support each other.
CRediT authorship contribution statement
After computing correlation matrix (Table 9), overall relationship
(Table 10) between CRs and ECs is determined. 𝐶𝑅1 and 𝐸𝐶 1 support
each other and change in 𝐸𝐶 1 also affects performance of a SNS with Akanksha Singh: Conceptualization, Data collection, Formal anal-
respect to 𝐶𝑅1 . Since any modification in 𝐸𝐶 2, 𝐸𝐶 3 and 𝐸𝐶 4 also ysis, Investigation, Methodology, Validation, Writing – original draft.
affects 𝐸𝐶 1 , therefore it is obligatory to compute overall relationship Sanjay Kumar: Conceptualization, Investigation, Methodology, Super-
between CRs and ECs. Based on overall relationship between CRs and vision, Validation, Writing – review & editing.
ECs and importance of CRs according to the users, relative ranking of
ECs is obtained as 𝐸𝐶 2 > 𝐸𝐶 3 > 𝐸𝐶 4 > 𝐸𝐶 5 > 𝐸𝐶 1 . The ECs with highest Declaration of competing interest
importance ratings are likely the features that a SNS should prioritize or
invest in more. Competitive assessment (Table 13) shows how different The authors declare that they have no known competing finan-
SNSs taken in the study perform with respect to different ECs. It is cial interests or personal relationships that could have appeared to
observed that performance of Facebook (𝐴1 ) with respect to all EC is
influence the work reported in this paper.
very satisfactory while Twitter (𝐴4 ) performs very good with respect to
𝐸𝐶 3 but has room for improvements with respect to other ECs. Finally,
performance of SNSs is aggregated (Table 14) and Facebook is ranked Acknowledgment
first with highest score value followed by Instagram, WhatsApp, and
Twitter. Research was supported by Council of Scientific & Industrial Re-
In the absence of parametric uncertainty, conventional QFD (Fig. 2) search (Grant File No. 09/171/ (0133)/2017-EMR-1) Government of
ranks Facebook as the best SNS followed by WhatsApp, Instagram and India under the scheme of Joint CSIR-UGC NET (JRF) and is gratefully
Twitter. When SNSs are analyzed using QFD and feedback from the acknowledged by first author.

Table 17
Linguistic rating of the alternatives with respect to CRs for validity test.
𝐷1 𝐷2 𝐷3 𝐷4
𝐴1 𝐴2 𝐴3 𝐴4 𝐴1 𝐴2 𝐴3 𝐴4 𝐴1 𝐴2 𝐴3 𝐴4 𝐴1 𝐴2 𝐴3 𝐴4
𝐶𝑅1 VG P VG MG MP MP VG P F MP VG F G MP VG F
𝐶𝑅2 G MG VP P G MP MP VP MG P F MP G G G MP
𝐶𝑅3 G P MP MG MG P G MP VG VP MP P MG F MG P
𝐶𝑅4 MG F P F VG P MG G VG MP P MG VG F MG G
𝐶𝑅5 F F G MG MP MP G F VP P VG MG MG MP G F

8
A. Singh and S. Kumar Engineering Applications of Artificial Intelligence 104 (2021) 104395

References Huarng, K., Yu, T.H.K., 2006. The application of neural networks to forecast fuzzy time
series. Physica A: Stat. Mech. Its Appli 363 (2), 481–491.
Abdul-Rahman, H., Kwan, C.L., Woods, P.C., 1999. Quality function deployment in Jian, S., Xiu-yan, P., Ying, X., Pei-Lei, W., Na-ji, M., 2016. A new method combin-
construction design: application in low-cost housing design. Int. J. Qual. Relia. ing QFD with intuitionistic fuzzy sets for web services selection. Int. J. Multi.
Manage 16 (6), 591–605. Ubiquitous Eng 11 (11), 107–118.
Aikhuele, D., 2017. Interval-valued intuitionistic fuzzy multi-criteria model for design Joshi, D., Kumar, S., 2018. An approach to multi-criteria decision making problems
concept selection. Manage. Sci. Letters 7 (9), 457–466. using dice similarity measure for picture fuzzy sets. In: Proceedings of International
Akao, Y., 1994. Development history of quality function deployment. Cus. Driven Appr. Conference on Mathematics and Computing. Springer, Singapore, pp. 135–140.
Qua. Plan. Deplo 339 (90). Karasu, S., Altan, A., Bekiros, S., Ahmad, W., 2020. A new forecasting model
Altan, A., Karasu, S., 2019. The effect of kernel values in support vector machine to with wrapper-based feature selection approach using multi-objective optimization
forecasting performance of financial time series. J. Cog. Sys 4 (1), 17–21. technique for chaotic crude oil time series. Energy 212, 118750.
Aslan, O., Altan, A., Hacioglu, R., 2017. The control of blast furnace top gas pressure by Khoo, L.P., Ho, N.C., 1996. Framework of a fuzzy quality function deployment system.
using fuzzy PID. In: Proceedings of the Fifth International Conference On Advances Int. J. Prod. Research 34 (2), 299–311.
In Mechanical And Robotics Engineering–AMRE, pp. 22-26. Kuo, I.H., Horng, S.J., Kao, T.W., Lin, T.L., Lee, C.L., Pan, L., 2009. An improved
Atanassov, K.T., 1999. Intuitionistic fuzzy sets. In: Intuitionistic Fuzzy Sets. Physica, method for forecasting enrollments based on fuzzy time series and particle swarm
Heidelberg, pp. 1–137. optimization. Expert Sys. Appli 36, 6108–6117.
Babar, A.H.K., Ali, Y., 2021. Enhancement of electric vehicles’ market competitiveness Lee, C.C., 1990a. Fuzzy logic in control systems: fuzzy logic controller I. IEEE Trans.
using fuzzy quality function deployment. Tech. Forecast. Social Change 167, Syst. Man Cybern. 20 (2), 404–418.
120738. Lee, C.C., 1990b. Fuzzy logic in control systems: fuzzy logic controller II. IEEE Trans.
Babbar, C., Amin, S.H., 2018. A multi-objective mathematical model integrating Syst. Man Cybern. 20 (2), 419–435.
environmental concerns for supplier selection and order allocation based on fuzzy Lee, A.W., Lin, G.T., Kuo, W.H., Lee, S.J., 2017. The application of quality function
QFD in beverages industry. Expert Syst. Appl 92, 27–38. deployment to smart-watches: The house of quality for improved product design. In:
Barnett, W.D., Raja, M.K., 1995. Application of QFD to the software development Proceedings of Portland International Conference on Management of Engineering
process. Int. J. Qual. Relia. Manage 12 (6), 24–42. and Technology (PICMET). IEEE, pp. 1–6.
Bevilacqua, M., Ciarapica, F.E., Giacchetta, G., 2006. A fuzzy-QFD approach to supplier Lee, G.H., Park, S.H., 2021. Fuzzy QFD-based prioritization of work activities of
selection. J. Purch. Supp. Manage 12 (1), 14–27. construction for safety. ICIC Exp. Leter Part B: App 12 (1), 1–8.
Cai, Q., Zhang, D., Wu, B., Leung, S.C., 2013. A novel stock forecasting model based Lee, C.K.M., Ru, C.T.Y., Yeung, C.L., Choy, K.L., Ip, W.H., 2015. Analyze the healthcare
on fuzzy time series and genetic algorithm. Proce. Comp. Sci 18, 1155–1162. service requirement using fuzzy QFD. Comp. Indus 74, 1–15.
Çalík, A., 2020. Evaluation of social media platforms using Best-Worst method and Li, M., Jin, L., Wang, J., 2014. A new MCDM method combining QFD with TOP-
fuzzy VIKOR methods: A case study of travel agency. Iran. J. Manage. Stu 13 (4), SIS for knowledge management system selection from the user’s perspective in
645–672. intuitionistic fuzzy environment. App. Soft Comput 21, 28–37.
Chen, S.M., 1996. A fuzzy reasoning approach for rule-based systems based on fuzzy Lima-Junior, F.R., Carpinetti, L.C.R., 2016. A multicriteria approach based on fuzzy QFD
logics. IEEE Trans. Syst. Man Cybern. B (Cybern) 26 (5), 769–778. for choosing criteria for supplier selection. Comput. Indus. Eng 101, 269–285.
Chen, S.M., Huang, C.M., 2003. Generating weighted fuzzy rules from relational Onar, S.Ç., Büyüközkan, G., Öztayşi, B., Kahraman, C., 2016. A new hesitant fuzzy QFD
database systems for estimating null values using genetic algorithms. IEEE Trans. approach: an application to computer workstation selection. App. Soft Comput 46,
Fuzzy Syst. 11 (4), 495–506. 1–16.
Chen, S.M., Jong, W.T., 1997. Fuzzy query translation for relational database systems. Quinn, A., Lemay, G., Larsen, P., Johnson, D.M., 2009. Service quality in higher
IEEE Trans. Syst. Man Cybern. B (Cybern) 27 (4), 714–721. education. Total Qual. Manage 20 (2), 139–152.
Chen, S.M., Kao, P.Y., 2013. TAIEX Forecasting based on fuzzy time series, particle Sakao, T., 2007. A QFD-centred design methodology for environmentally conscious
swarm optimization techniques and support vector machines. Info. Sci 247, 62–71. product design. Int. J. Prod. Research 45 (18–19), 4143–4162.
Chen, D.K., Wang, Y.Q., Jin, N., Liu, D.H., 2016. The application of QFD and Shi, H., Quan, M.Y., Liu, H.C., Duan, C.Y., 2018. A novel integrated approach for
information entropy in improved design for the manned capsule. In: Proceedings of green supplier selection with interval-valued intuitionistic uncertain linguistic
the 6th International Asia Conference on Industrial Engineering and Management information: A case study in the agri-food industry. Sustain 10 (3), 733.
Innovation, pp. 371-379. Singh, P., 2015. Correlation coefficients for picture fuzzy set. J. Int. Fuzzy Syst 28 (2),
Cuong, B.C., 2013a. Picture fuzzy sets-first results. Part 1. In: Neuro-Fuzzy Systems 591–604.
with Applications Seminar. Institute of Mathematics, Hanoi. Singh, A., Kumar, S., 2020. Picture fuzzy choquet integral-based VIKOR for multicriteria
Cuong, B.C., 2013b. Picture fuzzy sets-first results. Part 2. In: Neuro-Fuzzy Systems group decision-making problems. Gran. Comput. 1–15.
with Applications Seminar. Institute of Mathematics, Hanoi. Sireli, Y., Kauffmann, P., Ozan, E., 2007. Integration of Kano’s model into QFD for
Dat, L.Q., Phuong, T.T., Kao, H.P., Chou, S.Y., Van Nghia, P., 2015. A new integrated multiple product design. IEEE Trans. Eng. Mange 54 (2), 380–390.
fuzzy QFD approach for market segments evaluation and selection. Appl. Math. Song, Q., Chissom, B.S., 1993a. Fuzzy time series and its models. Fuzzy Sets Syst 54
Model. 39 (13), 3653–3665. (3), 269–277.
Dikmen, I., Birgonul, M.T., Kiziltas, S., 2005. Strategic use of quality function Song, Q., Chissom, B.S., 1993b. Forecasting enrollments with fuzzy time series – Part
deployment (QFD) in the construction industry. Build. Envir 40 (2), 245–255. I. Fuzzy Sets Syst 54 (1), 1–9.
Dong, J.J., Wu, J.Z., Pap, E., Szakal, A., 2017. A choquet capacity and integral based Tavana, M., Momeni, E., Rezaeiniya, N., Mirhedayatian, S.M., Rezaeiniya, H., 2013.
method to identify the overall importance of engineering characteristics in quality A novel hybrid social media platform selection model using fuzzy ANP and
function deployment. Eco. Comput. Eco. Cyber. Studies Res. 51 (4). COPRAS-G. Expert Syst. Appl. 40 (14), 5694–5702.
de Fátima Cardoso, J., Casarotto Filho, N., Miguel, P.A.C., 2015. Application of quality Terninko, J., 2018. Step-By-Step QFD: Customer-Driven Product Design. Routledge.
function deployment for the development of an organic product. Food Qual. Pref Wang, X., Triantaphyllou, E., 2008. Ranking irregularities when evaluating alternatives
40, 180–190. by using some ELECTRE methods. Omega 36 (1), 45–63.
Garg, H., 2017. Some picture fuzzy aggregation operators and their applications to Wei, G., 2017. Picture fuzzy aggregation operators and their application to multiple
multicriteria decision-making. Arab. J. Sci. Eng. 42 (12), 5275–5290. attribute decision making. J. Intell. Fuzzy Syst 33, 713–724.
Gitinavard, H., Mousavi, S.M., Vahdani, B., Siadat, A., 2016. A new distance-based Wei, G., 2018. Picture fuzzy hamacher aggregation operators and their application to
decision model in interval-valued hesitant fuzzy setting for industrial selection multiple attribute decision making. Funda. Info 157 (3), 271–320.
problems. Scientia Iran 23 (4), 1928–1940. Wu, S.M., Liu, H.C., Wang, L.E., 2017. Hesitant fuzzy integrated MCDM approach for
Haiyun, C., Zhixiong, H., Yüksel, S., Dinçer, H., 2021. Analysis of the innovation quality function deployment: a case study in electric vehicle. Int. J. Prod. Research
strategies for green supply chain management in the energy industry using the 55 (15), 4436–4449.
QFD-based hybrid interval valued intuitionistic fuzzy decision approach. Renew. Wu, W.Y., Qomariyah, A., Sa, N.T.T., Liao, Y., 2018. The integration between service
Sust. Energy Reviews 143, 110844. value and service recovery in the hospitality industry: An application of QFD and
Haktanír, E., Kahraman, C., Gündoğdu, F.K., 2021. Delivery drone design using ANP. Int. J. Hosp. Manage 75, 48–57.
spherical fuzzy quality function deployment. In: Proceedings of Decision Making Xu, Z., 2007. Intuitionistic fuzzy aggregation operators. IEEE Trans. Fuzzy Syst. 15 (6),
with Spherical Fuzzy Sets. Springer, Cham, pp. 399–430. 1179–1187.
Hauser, J.R., Clausing, D., 1988. The house of quality. Zadeh, L.A., 1965. Fuzzy sets. Info. Cont 8 (3), 338–353.
Herzwurm, G., Schockert, S., Tauterat, T., 2015. Quality function deployment in Zeng, S., Qiyas, M., Arif, M., Mahmood, T., 2019. Extended version of linguistic picture
software development-state-of-the-art. In: Proceedings of the 21th International fuzzy TOPSIS method and its applications in enterprise resource planning systems.
Symposium on Quality Function Deployment, pp. 1-18. Math. Prob. 8594938.
Zhong, J., Guo, A., Fu, Y., Wang, H., 2018. Application of QFD in China’s higher
education: A bibliometric study. In: Proceedings of 4th International Conference on
Economics, Social Science, Arts, Education and Management Engineering (ESSAEME
2018). Atlantis Press, pp. 249–254.

You might also like