Content uploaded by Xuening Chu
Author content
All content in this area was uploaded by Xuening Chu on Jan 14, 2021
Content may be subject to copyright.
A new importance–performance analysis approach for customer satisfaction
evaluation supporting PSS design
Xiuli Geng, Xuening Chu
⇑
School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200240, PR China
article info
Keywords:
Product-service system (PSS)
Importance–performance analysis (IPA)
Kano’s model
Decision making trial and evaluation
laboratory (DEMATEL)
Vague sets
abstract
Product-service system (PSS) design focuses on customer value and satisfaction more than traditional
product or service design, and pays much attention to making improvement strategies due to the imma-
turity of engineering design methodology. Customer satisfaction evaluation attracts PSS providers’ atten-
tions in supporting PSS design. Importance–performance analysis (IPA) as an effective customer
satisfaction evaluation tool is revised and used to identify PSS improvement strategies in this paper.
The new IPA is proposed for three reasons. First, considering the fact that the attribute performance
and importance are not independent variables and attribute performance has a nonlinear relationship
with the overall satisfaction, Kano’s model is integrated into IPA. Second, to overcome the drawbacks
of statistic method and artificial neural network (ANN) in obtaining attribute importance implicitly,
e.g. requiring sufficient and confident data, and overlooking the attribute original importance about attri-
bute’s contributing level to customer value realization, a set of adjustment models are proposed to revise
the attributes original importance according to the Kano quality categories of attributes and the levels of
attributes performance. Third, considering the mutual influence relationships among attributes, the pro-
posed IPA takes these relationships into account by decision making trial and evaluation laboratory
(DEMATEL). In addition, to deal with the uncertainty and vagueness in evaluation process, vague sets
are employed in the revised IPA. A case study is carried out to demonstrate the effectiveness of the devel-
oped customer satisfaction evaluation approach.
Ó2011 Elsevier Ltd. All rights reserved.
1. Introduction
The emerging concept of product-service system (PSS) has the
potential to drive sustainable production and consumption, and
expand companies’ competitive space and customers’ satisfaction
space. For the manufacturing industry, the traditional boundary
between manufacturing and services is becoming increasing
blurred (Mont, 2002). The new trend is products and services are
integrated and provided as a whole set to fulfill customer’s
requirements, and the product/service ratio can vary in different
customer using contexts. The driving force of PSS design is not tra-
ditional functional requirements for product design or service de-
sign, but a higher requirement level named customer value. Sakao,
Shimomura, Sundin, and Comstock (2009) proposed that the target
in service product engineering (SPE) was shifted from functions or
quality to value, and developed a service model consisting four
sub-models: ‘‘flow model’’, ‘‘scope model’’, ‘‘scenario model’’, and
‘‘view model’’.
Continuously satisfying customer value requirements is a
critical strategy for PSS design. Therefore, analyzing customer
satisfaction and identifying improvement opportunities are impor-
tant tasks for PSS providers. Furthermore, the PSS engineering de-
sign methodologies are still immature, and it makes the
continuous improvement in PSS design much more important.
Importance–performance analysis (IPA) introduced by Martilla
and James (1977) is a simple and effective technique for customer
satisfaction evaluation, which can assist practitioners in identify-
ing attributes improvement priorities and making quality-based
marketing strategies to achieve advantages over competitors
(Abalo, Varela, & Manzano, 2007; Hansen & Bush, 1999). Its main
structure is dividing attributes into four groups depending on their
performance and importance to customers.
Conventional IPA model has two impractical assumptions
(Matzler, Bailom, Hinterhuber, Renzl, & Pichler, 2004): (1) the rela-
tionship between attribute performance and overall customer sat-
isfaction is linear, (2) attribute importance and attribute
performance are independent variables. For the first assumption,
the most famous model to describe that the relationship between
attribute performance and overall customer satisfaction is not
always linear is the Kano’s model proposed by Kano, Seraku,
Takahashi, and Tsuji (1984). For the second assumption, the fact
that the attribute performance and importance are related has also
been suggested (Matzler et al., 2004; Ryan & Huyton, 2002). In
0957-4174/$ - see front matter Ó2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2011.08.038
⇑
Corresponding author. Tel./fax: +86 21 3420 6339.
E-mail address: xnchu@sjtu.edu.cn (X. Chu).
Expert Systems with Applications 39 (2012) 1492–1502
Contents lists available at SciVerse ScienceDirect
Expert Systems with Applications
journal homepage: www.elsevier.com/locate/eswa
order to overcome the erroneous assumptions in the traditional
IPA model, many revised IPA models have been presented. Their
main focuses are revising the method of eliciting attribute impor-
tance, because the customer self-stated attribute importance is
insufficient for effective IPA analysis.
Two categories of approaches have been used to acquire the
attribute importance implicitly: statistical approach and artificial
neural network (ANN). Matzler and Sauerwein (2002) applied the
multiple regression analysis to correlate the attribute performance
with overall satisfaction, and used the regression coefficients as the
attributes importance. Matzler, Sauerwein, and Heischmidt (2003)
used the partial ranking method to derive partial correlation coef-
ficients as attributes importance. These conventional statistic
methods have some drawbacks, e.g. they assume that the attribute
performance has a linear relationship with the overall satisfaction.
Garver (2002) indicated that ANN can be employed to overcome
the constraint of hypothetical problems of conventional statistic
method. Deng, Chen, and Pei (2008) used the back-propagation
neural network (BPNN) model to establish the function relation-
ships between attributes performance and the integral satisfaction
and derive the hidden importance value of each attribute. Deng
(2007) integrated the three-factor theory and partial correlation
analysis to amend the IPA model considering the nonlinear rela-
tionships between attributes performance and the overall satisfac-
tion. These researches have not considered the mutual influence
relationships among attributes in IPA, and these relationships have
a significant effect on the rationality of analyzing outcome. Hu, Lee,
and Yen (2009) integrated BPNN and decision making trial and
evaluation laboratory (DEMATEL) to re-establish the IPA model,
and DEMATEL was used to analyze the causal relationship among
the quality characteristics. The DEMATEL method, originated from
the Geneva Research Center of the Battelle Memorial Institute
(Gabus & Fontela, 1972, 1973), is especially practical for analyzing
influence relationships among elements in a system, and identify-
ing causal group and effect group in the system through a causal
diagram.
The revised IPA models based on statistic approaches or ANN
have two weak points: (1) the application and success of them de-
pend on sufficient and confidential historical data, (2) the attribute
importance does not embody the basic importance of attribute to
customer value. In the setting of this paper, historical data are defi-
cient in the early phase in the PSS companies, thus the statistic ap-
proaches and intelligent approaches lack effectiveness in deriving
attributes importance. Hu, Lee, and Yen (2009) and Hu, Lee, Yen,
and Tsai (2009b) proposed a KD-IPA method to obtain attribute
importance through modifying the original attribute importance
based on Kano’s model and DEMATEL. However, the unavoidable
uncertainty and vagueness in the evaluation process is not consid-
ered in KD-IPA.
It is more rational for customers to adopt imprecise linguistic
terms to express their judgments in IPA. Therefore, how to deal
with linguistic information is critical to the effectiveness of evalu-
ation. Fuzzy set theory is commonly used in dealing with linguistic
terms. However, it supports the favoring evidences only, and just
reveals the positive preference degree associated with decision
maker’s (DM’s) subjective judgments. A single membership degree
provides little information about its uncertainty. Consequently,
fuzzy set theory has some shortcomings in dealing with vagueness
information. Recent years, vague set theory proposed by Gau and
Buehrer (1993) gains its popular in decision-making in many areas,
e.g. idea-screening in new product development (Lo, Wang, &
Chao, 2006), engineering schemes selection (Ye, 2007), and sup-
plier selection (Boran, Gen, Kurt, & Akay, 2009; Zhang, Zhang, Lai,
& Lu, 2009). Vague set theory supports both the favoring and
opposing evidences in dealing with DMs’ linguistic judgments by
means of membership function and non-membership function.
Therefore, it is superior to fuzzy set theory in capturing uncertainty
and vagueness that exists in IPA analysis.
This paper proposes a new IPA approach based on vague sets for
customer satisfaction evaluation. Kano’ model and DEMATEL are
employed in the new IPA to consider the nonlinear impact of PSS
quality attributes and causal relationship among these attributes,
respectively. The original attribute importance is acquired from
customers to express the importance of this attribute contributing
to customer value requirements. At the same time, the original
attribute importance can reflect the attribute expectation level to
some degree. The attribute original importance is modified twice.
First, it is modified according to the Kano quality category the attri-
bute belongs to and the level of the attribute performance. Three
adjustment models for modifying attribute original importance
are proposed aiming at three different Kano quality categories,
and these models are based on the comparison and similarity cal-
culations between attribute performance and attribute original
importance. Second, it is modified by the difference level of the
attribute impacting others and being impacted by others driving
from DEMATEL. The remainder of the paper consists of the follow-
ing sections. Section 2gives out a description of the proposed prob-
lem and approach, and reviews the relative literature. Section 3
gives out a description of the first modification of attribute original
importance based on Kano’s model and vague sets. Fuzzy Kano
analysis to determine the Kano quality category of attribute and
the three adjustment models are presented. Section 4presents
the second modification of attribute importance based on
DEMATEL with vague sets. In Section 5, the proposed approach is
applied in a real world case of customer satisfaction evaluation
supporting PSS design for a company providing pump products
and services. Conclusions are then presented in Section 6.
2. Problem description and literature review
The researches on PSS design are immature and on the way of
searching. Most of the early studies on PSS design were primarily
conducted from the viewpoint of marketing and management, such
as case study approach (Manzini and Vezzoli, 2003) and industry
PSS initiatives research at the empirical level (Williams, 2007).
Engineering methods and tools have been developed gradually to
support the realization of PSSs, such as life-cycle engineering for
technical PSS development (Aurich, Fuchs, & DeVries, 2004; Aurich
et al., 2006), PSS representation approach based on IDEF0 (Morelli,
2006), and service engineering based on computer-aided design
tool (Sakao & Shimomura, 2007; Sakao et al., 2009). Satisfying cus-
tomer is the design objective pursued unceasing for PSS providers.
Evaluating customer satisfaction to identify improvement opportu-
nities of PSS design gets much more important in practice.
Extracting critical customer perception attributes of PSS is the
fundamental step in evaluating customer satisfaction. Question-
naire survey is an effective way to collect objective information.
The quality characteristics generated in the PSS conceptual design
can be combined with the data in customer relationship manage-
ment in offering fundamental information to design questionnaire
for eliciting critical customer perception attributes. After acquiring
customer perception PSS attributes, the rest task is identifying the
management strategies of these attributes based on IPA.
IPA is a two-dimensional grid based on customer-perceived
importance and performance of attribute analyzed. The x-axis
and y-axis presents attribute performance and attribute impor-
tance, respectively. These two axes divide the IPA grid into four
quadrants. The graphic representation provides an understandable
guide for identifying the crucial product or service attributes in
terms of their need for managerial action. IPA has been widely used
in customer satisfaction analysis for improving service quality in
many fields, e.g., tourism management (Enright & Newton, 2004;
X. Geng, X. Chu / Expert Systems with Applications 39 (2012) 1492–1502 1493
Huan, Beaman, & Shelby, 2002; Zhang & Chow, 2004), park envi-
ronmental management (Tonge & Moore, 2007), bank service
(Matzler et al., 2003), and hotel industry (Chu & Choi, 2000).
How to evaluate attribute importance and attribute performance
is critical important to the effectiveness of IPA. The attribute per-
formance and attribute original importance can be evaluated di-
rectly by customers in linguistic variables based on their
experiences. The granularity of linguistic judgments of attribute
performance and original importance and the corresponding vague
set value scales are the same, thus it can be ensured that the attri-
bute performance and original importance can be compared.
The actual attribute importance is affected by the attribute per-
formance. Considering the linear or nonlinear relationship be-
tween attribute performance and overall satisfaction, the original
importance of attributes, which belong to different Kano quality
categories, are revised by different adjustment models based on
the deviation level between attribute original importance and
attribute performance. The two-dimensional Kano quality model
and Kano evaluation table (Matzler & Hinterhuber, 1998) are tradi-
tional tools to characterize product or service attributes. In Kano’s
model, attributes are classified into five categories: must-be attri-
bute (M), one-dimensional attribute (O), attractive attribute (A),
indifferent attribute (I), and reverse attribute (R). The traditional
Kano’s questionnaire method uses binary data to model the opin-
ions of customers and/or engineers who are asked to complete
the questionnaires. However, opinions of individual persons are
subjective and vague due to their inadequate information and
experiences. Fuzzy questionnaire can provide opportunities for
the respondents to express their feelings using subjective and va-
gue descriptions. Lee and Huang (2009) developed fuzzy Kano’s
questionnaire (FKQ) and fuzzy Kano’s mode (FKM) for analyzing
the customer requirements.
Considering the cause-effect relationships among attributes
analyzed, the attribute importance need to be revised further.
DEMATEL has been successfully applied to construct the causal
relations among the elements in a system in many situations such
as marketing strategies (Chen, Lien, Tzeng, & Yang, 2009), e-learn-
ing program (Tzeng, Chiang, & Li, 2007), and safety problems (Ou
Yang, Shieh, Leu, & Tzeng, 2008). Lee, Li, Yen, and Huang (2010)
adopted DEMATEL to analyze the interaction influence level be-
tween technology acceptance model (TAM) variables, re-establish
the TAM structure and grasp the key variables of TAM to make a
better resources deployment. Wu (2008), Chen and Chen (2010),
and Chen, Lien, and Tzeng (2010) combined DEMATEL with ana-
lytic network process (ANP) for supporting decision-making. DEM-
ATEL is used to construct an interaction relationships network
among criteria for implementing ANP. In order to deal with the is-
sue that the crisp numbers are hard to evaluate the influence rela-
tionships within the DEMATEL matrix effectively, fuzzy theory has
been integrated in DEMATEL application (Lin & Wu, 2008). There is
no research on integrating vague sets in DEMATEL. The framework
of the proposed approach for customer satisfaction evaluation sup-
porting PSS design is shown in Fig. 1.
3. The first modification of attribute original importance based
on Kano’s model and vague sets
3.1. Vague sets
The basic concept of vague set theory is as follows. Let a set
X={x
1
,x
2
,...,x
n
} be a finite universal set. A vague set Von Xis
an object with the form: V={(x,[t
V
(x), 1 f
V
(x)])|x
e
V}, where the
functions: t
V
:X?[0, 1] and f
V
:X?[0, 1]. A vague set Vis
characterized by a truth-membership function t
V
and a false-mem-
bership function f
V
, where 0 6t
V
(x)+f
V
(x)61.
Definition 1. Zhang et al. (2009) gave out the definition of
comparison between vague sets according to the characteristics
of interval-value numbers.
For vague value x=[t
x
,1f
x
], y=[t
y
,1f
y
], the probability of
xPyis defined by
pðxPyÞ¼Max 0;LðxÞþLðyÞMaxð0;1f
y
t
x
Þ
LðxÞþLðyÞ;ð1Þ
where L(x)=1f
x
t
x
,L(y)=1f
y
t
y
is the length of vague value
x,y.
Definition 2. Zhang, Huang, and Li (2004) proposed a method to
calculate the similarity measure between vague values
x=[t
x
,1f
x
], y=[t
y
,1f
y
]:
Sðx;yÞ¼1dðx;yÞ
ffiffiffi
2
p;ð2Þ
where
dðx;yÞ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðt
x
t
y
Þ
2
þð1f
x
ð1f
y
ÞÞ
2
q¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðt
x
t
y
Þ
2
þðf
x
f
y
Þ
2
qis
the distance between xand y.
Definition 3. For vague value x=[t
x
,1f
x
], the defuzzificated
value of xis defined as follows (Zhang et al., 2009):
Df ðxÞ¼t
x
=t
x
þf
x
ðÞ:ð3Þ
3.2. Determining the Kano quality categories of attributes using fuzzy
Kano analysis
The customer perception PSS quality attributes are identified
through questionnaire survey. The two-dimensional Kano ques-
tionnaire uses functional and dysfunctional models to ask design
Fig. 1. The framework of the proposed approach for customer satisfaction
evaluation and management.
1494 X. Geng, X. Chu / Expert Systems with Applications 39 (2012) 1492–1502
engineers and customers to think how potential customers would
feel if an attribute is either presented or not. Linguistic variables,
such as ‘‘like’’, ‘‘must-be’’, ‘‘neutral’’, ‘‘live-with’’ and ‘‘dislike’’, are
used to answer the functional questions and dysfunctional ques-
tions respectively. These linguistic variables are assigned with fuz-
zy membership functions. An example of FKQ is shown in Table 1.
FKM (Lee & Huang, 2009), as a fuzzy statistics method, can be
applied to collect the information in FKQs. Each attribute can be
categorized into one of the five Kano’s attribute classes based on
the Kano evaluation table given in Table 2.
Let Uand Vbe the universal sets of functional and dysfunctional
questions, P={P
1
,P
2
,...,P
m
} and N={N
1
,N
2
,...,N
n
} be the sets
with mand nlinguistic variables on Uand V, respectively, which
jointly construct an mnevaluation matrix S
k
of the two-dimen-
sional quality model for the kth person. Let rbe the total number of
persons to be interviewed. For each person, the linguistic variables
m(P)
ki
and m(N)
kj
are normalized to satisfy P
m
i¼1
mðPÞ
ki
¼1 and
P
n
j¼1
mðNÞ
kj
¼1.
Let S
k
ij
¼mðPÞ
ki
mðNÞ
kj
to achieve an mby nmatrix:
In the FKM used in this research, P=N= {Like, Must-be, Neu-
tral, Live-with, Dislike}, and m=n= 5. Let T
k
h
¼PS
k
xy
ðh¼M;O;
A;I;R;QÞ, where the position of (x,y) cell represents the hclass of
Kano’s attribute in the evaluation sheet. T
k
h
reflects to what extent
the EC belongs to the hclass. Given a classification threshold
a
,if
T
k
h
P
a
, then T
k
h
¼1, otherwise T
k
h
¼0.
For each attribute, calculate T
k
h
for k¼1;2;...;r, and count the
number when T
k
h
¼1 is satisfied for each of the attribute classes:
F
a
fhg¼fM;O;A;I;R;Qg. The attribute belongs to the hattribute
class which corresponds to the maximum value of F
a
fhg. For exam-
ple, when F
a
¼0:5
fM;O;A;I;R;Qg¼f22;16;12;0;0;0gis obtained for
one attribute, this attribute belongs to the ‘‘must-be (M)’’ attribute
class.
Among the five categories of Kano quality attributes, must-be
attribute, one-dimension attribute and attractive attribute are
three common categories and have research value. Therefore, if a
PSS quality attribute belongs to indifferent attribute or reverse
attribute, it is not analyzed in the proposed IPA.
3.2.1. Must-be attribute
Must-be attribute is the basic quality characteristic considered
by customers. If the attribute performance is better than custom-
ers’ expectation, customers’ satisfaction level will not increase a
lot, and customers will pay less attention to the attribute. In other
words, the attribute original importance will decrease a lot. Other-
wise, if the attribute performance is worse than customers’ expec-
tation, customers’ satisfaction level will decrease dramatically, and
customers will pay much more attention to the attribute. In other
words, the attribute original importance will increase a lot.
3.2.2. One-dimension attribute
The performance level of the one-dimension attribute is propor-
tional to the customer satisfaction level. If the attribute perfor-
mance is better than customers’ expectation, customers’
satisfaction level will increase accordingly. In other words, the
attribute original importance will decrease accordingly. Otherwise,
the attribute original importance will increase accordingly. The dif-
ference level between the attribute performance and customers’
expectation is proportional to the decrease or increase scale of
the attribute original importance.
3.2.3. Attractive attribute
Usually, customers have low expectation to the attractive attri-
bute. If the attribute performance is better than the customers’
expectation, customers will feel very excited. However, they will
not lose much attention to the attribute. In other words, the attri-
bute original importance will decrease a little. Otherwise, the attri-
bute original importance will increase a little.
3.3. The adjustment models of attribute original importance based on
Kano’s model and vague sets
Survey questionnaires are sent to customers to elicit their lin-
guistic judgments on attributes performance and attributes origi-
nal importance. These linguistic variables can be expressed in
vague values on a 1–7 scales, as shown in Table 3. Assume
A={A
1
,A
2
,...,A
n
} is the set of ncustomer perceived quality attri-
butes; C={C
1
,C
2
,...,C
h
} is the set of hcustomers surveyed. The
performance and the original importance of attribute A
i
(1 6i6n)
given by customer C
k
(1 6k6h) is represented as ^
P
k
i
and ^
I
k
i
,
respectively.
The aggregated judgments of all the customers on attribute per-
formance ^
P
i
and original importance ^
I
i
can be acquired by averag-
ing method:
^
P
i
¼1
hX
h
k¼1
^
P
k
i
;ð4Þ
^
I
i
¼1
hX
h
k¼1
^
I
k
i
:ð5Þ
Table 1
The fuzzy Kano questionnaire.
Like Must-be Neutral Live-with Dislike
Functional m
1
m
2
00 0
Dysfunctional 0 0 m
3
m
4
m
5
Table 2
Kano evaluation table.
Functional Dysfunctional
Like Must-be Neutral Live-with Dislike
Like QA A A O
Must-be RI I I M
Neutral RI I I M
Live-with RI I I M
Dislike RR R R Q
Table 3
The scale of linguistic variables for performance and importance.
Attribute performance Original attribute importance Vague value
Very poor (VP) Very low (VL) [0.0, 0.1]
Poor (P) Low (L) [0.1, 0.3]
Medium poor (MP) Medium low (ML) [0.3, 0.4]
Medium (M) Medium (M) [0.4, 0.5]
Medium good (MG) Medium high (MH) [0.5, 0.6]
Good (G) High (H) [0.6, 0.9]
Very good (VG) Very high (VH) [0.9, 1.0]
X. Geng, X. Chu / Expert Systems with Applications 39 (2012) 1492–1502 1495
^
P
i
and ^
I
i
can be defuzzificated by Eq. (3) to get the precise value
P
i
and I
i
. Three different adjustment models are proposed for revis-
ing the original importance of three different categories of attri-
butes. If the attribute performance is better than the customer
expected performance (attribute original importance), e.g.
pð^
P
i
P^
I
i
ÞP0:5, the original importance of the attribute will de-
crease. For attributes belonging to different Kano quality catego-
ries, the original importance decreasing scales of them are
different. Otherwise, if the attribute performance is worse than
the customer expected performance (attribute original impor-
tance), e.g. pð^
P
i
P^
I
i
Þ<0:5, the original importance of the attribute
will increase. ð1Sð^
P
i
;^
I
i
ÞÞ reflects the difference levels between
the attribute performance and customer expected performance.
The original importance of must-be Kano quality attribute is more
sensitive to the difference level than that of the one-dimension
Kano quality attribute, and original importance of the one-dimen-
sion Kano quality attribute is more sensitive to the difference level
than that of the attractive Kano quality attribute .The three adjust-
ment models of attribute original importance taking into account
attribute performance level and Kano quality category are as
follows:
Model 1. When the attribute is a must-be Kano quality attribute,
the original importance adjustment model is as follows:
I
0
i
¼
I
i
11S^
P
i
;^
I
i
hi
2
if p^
P
i
P^
I
i
P0:5;
I
i
1þ1S^
P
i
;^
I
i
hi
2
otherwise p^
P
i
P^
I
i
<0:5:
8
>
<
>
:ð6Þ
Model 2. When the attribute is a one-dimension Kano quality
attribute, the original importance adjustment model is as follows:
I
0
i
¼
I
i
11S^
P
i
;^
I
i
hi
if p^
P
i
P^
I
i
P0:5;
I
i
1þ1S^
P
i
;^
I
i
hi
otherwise p^
P
i
P^
I
i
<0:5:
8
>
<
>
:ð7Þ
Model 3. When the attribute is an attractive Kano quality attribute,
the original importance adjustment model is as follows:
I
0
i
¼
I
i
11S^
P
i
;^
I
i
hi
1=2
if p^
P
i
P^
I
i
P0:5;
I
i
1þ1S^
P
i
;^
I
i
hi
1=2
otherwise p^
P
i
P^
I
i
<0:5:
8
>
<
>
:ð8Þ
4. The second modification of attribute importance based on
DEMATEL
There exist cause-effect relationships among the derived attri-
butes for analysis. DEMATEL is employed to determine the cause-
effect model of these attributes. It is based on evaluating the direct
influence relationships among the attributes group by experts and
professional customers. Vague sets are used to express the DMs’
evaluation to improve the rationality of the decision-making
process.
4.1. DEMATEL based on vague sets
The procedures of DEMATEL analysis based on vague sets are as
follows:
Step 1. Establish a direct-relation matrix.
Suppose the attributes set for DEMATEL analysis is A=
{A
1
,A
2
,...,A
n
}. Questionnaire survey is conducted on experts
and professional customers to elicit the linguistic judgments
on the direct influence relationships between two attributes.
The linguistic judgments can be expressed in vague numbers
according to Table 4.
Suppose M={M
1
,M
2
,...,M
g
} is the set of DM group.
^
x
k
ij
ð16k6gÞis the evaluation in vague number about the level
of attribute A
i
impacting attribute A
j
given by the DM k, then the
direct-relation matrix given by DM k,^
X
k
,ofnncan be ob-
tained. The diagonal variable ^
x
k
ii
of the direct-relation matrix is
set to 0:
^
X
k
¼
0^
x
k
12
^
x
k
1n
^
x
k
21
0 ^
x
k
2n
.
.
..
.
...
..
.
.
^
x
k
n1
^
x
k
n2
0
2
6
6
6
6
6
4
3
7
7
7
7
7
5
;k¼1;2;...;g;
where ^
x
k
ij
¼½x
k
ijl
;x
k
iju
.
Step 2. Calculate normalized direct-relation matrix.
The common method of normalizing the direct-relation matrix
is based on the biggest sum of the row vectors as the normal-
ized base (Kim, Park, & Lee, 2007; Lin & Wu, 2008; Seyed-Hos-
seini, Safaei, & Asgharpour, 2006; Wu & Lee, 2007). Set the
normalized coefficient for ^
X
k
is k
k
. The normalized direct-rela-
tion matrix ^
Z
k
for ^
X
k
is calculated as ^
Z
k
¼k
k
^
X
k
:
k
k
¼1
max
16i6n
P
n
j¼1
x
k
iju
:ð9Þ
All the normalized direct-relation matrices according to differ-
ent DMs’ opinions are integrated into a consensus one ^
Z:
^
Z¼
0^
z
12
^
z
1n
^
z
21
0 ^
z
2n
.
.
..
.
...
..
.
.
^
z
n1
^
z
n2
0
2
6
6
6
6
4
3
7
7
7
7
5
;where ^
z
ij
¼z
ijl
;z
iju
;
z
ijl
¼P
p
k¼1
z
k
ijl
g;z
iju
¼P
p
k¼1
z
k
iju
g:
Step 3. Calculate the total-relation matrix.
If the normalized direct-relation matrix Zis crisp, the derived
total-relation matrix Tcan be obtained from Eq. (10) (Huang
et al., 2007):
T¼lim
w!1
ZþZ
2
þþZ
w
¼ZðIZÞ
1
;ð10Þ
where Iis an identity matrix.
How to calculate the vague total-relation matrix ^
Tfrom the va-
gue normalized direct-relation matrix ^
Zhas to be solved. To
compute the total-relation matrix ^
T, the convergence of
lim
w!1
^
Z
w
¼0 has to be ensured:
Table 4
The scale of linguistic variables for strength of relationship.
Influencing levels Vague value
Very weak (VW) [0.0, 0.1]
Weak (W) [0.1, 0.3]
Medium weak (MW) [0.3, 0.4]
Medium (M) [0.4, 0.5]
Medium strong (MS) [0.5, 0.6]
Strong (S) [0.6, 0.9]
Very strong (VS) [0.9, 1.0]
1496 X. Geng, X. Chu / Expert Systems with Applications 39 (2012) 1492–1502
Let ^
Z
w
¼
0^
z
w
12
^
z
w
1n
^
z
w
21
0 ^
z
w
2n
.
.
..
.
...
..
.
.
^
z
w
n1
^
z
w
n2
0
2
6
6
6
6
4
3
7
7
7
7
5
;where ^
z
w
ij
¼z
w
ijl
;z
w
iju
hi
:
Define Z
w
l
¼z
w
ijl
hi
¼
0z
w
12l
z
w
1nl
z
w
21l
0 z
w
2nl
.
.
..
.
...
..
.
.
z
w
n1l
z
w
n2l
0
2
6
6
6
6
4
3
7
7
7
7
5
;
Z
w
u
¼z
w
iju
hi
¼
0z
w
12u
z
w
1nu
z
w
21u
0 z
w
2nu
.
.
..
.
...
..
.
.
z
w
n1u
z
w
n2u
0
2
6
6
6
6
4
3
7
7
7
7
5
:
Since Z
w
l
and Z
w
u
are crisp matrices, Eqs. (11) and (12) can be verified
according to the research of Lin and Wu (2008):
lim
w!1
IþZ
l
þZ
2
l
þþZ
w
l
¼IZ
l
ðÞ
1
ð11Þ
lim
w!1
IþZ
u
þZ
2
u
þþZ
w
u
¼IZ
u
ðÞ
1
:ð12Þ
Define vague total-relation matrix:
^
T;^
T¼
0^
t
12
^
t
1n
^
t
21
0 ^
t
2n
.
.
..
.
...
..
.
.
^
t
n1
^
t
n2
0
2
6
6
6
6
4
3
7
7
7
7
5
;where ^
t
ij
¼t
ijl
;t
iju
:
According to Eq. (10), matrix:
t
ijl
¼lim
w!1
Z
l
þZ
2
l
þþZ
w
l
¼Z
l
IZ
l
ðÞ
1
:ð13Þ
Similarly, matrix:
t
iju
¼Z
u
IZ
u
ðÞ
1
:ð14Þ
Now that the vague total-relation matrix ^
Tis obtained, define ^
D
i
is
the sum of the ith row of ^
Tand ^
R
j
is the sum of the jth column of ^
T,
and they are easy to be calculated. ^
D
i
summarizes both direct and
indirect influences given by attribute A
i
to others. ^
R
j
denotes both
direct and indirect influences received by attribute A
j
from others.
^
D
i
and ^
R
j
are interval numbers in fact, and they can be transformed
into crisp values using simple method as Eqs. (15) and (16):
D
i
¼D
il
þD
iu
ðÞ
2;where ^
D
i
¼D
il
;D
iu
½;ð15Þ
R
i
¼R
il
þR
iu
ðÞ
2;where ^
R
i
¼R
il
;R
iu
½:ð16Þ
Step 4. Determine the cause group and effect group.
D
i
+R
i
shows the overall level of attribute A
i
being influenced
and the influence on other attributes. It reflects the prominence
of attribute A
i
in the system. D
i
R
i
shows the difference level of
attribute A
i
being influenced and the influence to others. It de-
picts the net influence contributing to the whole system. If it
is positive, attribute A
i
belongs to the cause group. If it is nega-
tive, attribute A
i
belongs to the effect group.
4.2. IPA analysis
The final attribute importance I
r
i
is obtained by taking into ac-
count the mutual influence relationships among the attributes
group. If an attribute has higher net influence contributing to the
whole system, it is more important:
I
r
i
¼I
0
i
þðD
i
R
i
Þ:ð17Þ
The final procedure of the IPA approach is drawing the IPA map
and analyzing quality-based management strategies. In IPA, the x-
axis and y-axis present attribute performance and attribute impor-
tance, respectively. These two axes divide the IPA grid into four
quadrants, as shown in Fig. 2. Attributes falling in different quad-
rants need different management strategies. Attributes falling in
Quadrant I have both high performance and importance. It indi-
cates that the PSS solution has been performing well on the attri-
butes in this quadrant to gain competitive advantage. Attributes
falling in Quadrant II have high performance but low priority,
which have been overemphasized. Attributes falling in Quadrant
III have the characteristics of both low performance and impor-
tance, which can be considered as the weakness of the PSS. Attri-
butes falling in Quadrant IV have low performance but high
importance, which require immediate attention for improvement.
5. Case study
As one of the world’s top 500 companies, company H manufac-
tures world-class metering pumps and provides services to their
pump products in the Asia Pacific region. The main products of
Fig. 2. Importance–performance analysis.
Table 5
PSS quality attributes elicited from questionnaire survey.
Attributes
A1 Flow adjustment level
A2 Flow accuracy
A3 Self-monitoring ability
A4 Component reliability
A5 Energy saving and environmental friendly level
A6 Service response time
A7 Efficiency of field service
A8 Efficiency of equipment return maintenance
A9 Reliability of preventive maintenance
A10 Technical supporting extent
A11 Training level aiming at operating
A12 Professional level of service staff
Table 6
The FKQ of respondent 1 with respect to A6.
Like Must-be Neutral Live-with Dislike
Functional 0.6 0.4 0 0 0
Dysfunctional 0 0 0 0.3 0.7
X. Geng, X. Chu / Expert Systems with Applications 39 (2012) 1492–1502 1497
its Chinese subsidiary are the reciprocating metering pumps and
the leak-proof centrifugal pumps, which are widely used in waste
water treatment, petrochemical and chemical industry, pharma-
ceutical and food industry, etc. Although a reciprocating metering
pump is not used alone, it serves as the key module in the whole
system. The function of a metering pump is to dose chemical medi-
cament into the system at specific time and with specific quantity.
Its operating stability and robustness have direct impact on the
quality of the system output such as the composition of the treated
waste water.
Waste water treatment is important in municipal management.
Among various water treatment methods, the chemical treatment
method is the most popular one. Dosing chemical medicament is a
critical task in the process of chemical treatment method. The
reciprocating metering pump is used to treat the waste water by
dosing polymer into it. Customer value requirement is a stable
and reliable polymer dosing function. Customers pay attention to
not only the dosing function of the machine, but also the mainte-
nance of the required function in its lifecycle. The PSS development
needs to consider both the pump and its related technical services
to satisfy the individual customer requirement and enhance the
company’s competitiveness. Customer satisfaction should be eval-
uated aiming at different customer groups, so improvement oppor-
tunities can be identified and customer satisfaction management
strategies can be made. The proposed IPA approach is used for cus-
tomer satisfaction evaluation supporting PSS design.
Survey questionnaire to elicit customer perception PSS quality
attributes is designed by experts according to the PSS conceptual
design information and information coming from customer rela-
tionship management. Survey questionnaires are sent to 15 expe-
rienced customers to obtain the major PSS quality attributes. 12
attributes are elicited as shown in Table 5.
5.1. Determining the kano quality category of each attribute
FKQs are designed and distributed to various customers and
engineers. In this research, 50 effective samples are collected.
The functional question in FKQ with respect to A6 is ‘‘If the service
requirement you asked to the call center is answered within 24 h,
how you would feel?’’ The dysfunctional question for A6 is ‘‘If the
service requirement you asked to the call center is not answered
within 24 h, how you would feel?’’ The answers of the respondent
1(FS
1
) with respect to A6 are shown in Table 6.m=n= 5. FKM is
then used to obtain and analyze the total attribute level of A6:
mðPÞ
15
¼f0:6;0:4;0;0;0g;
mðNÞ
15
¼f0;0;0;0:3;0:7g:
Table 10
The judgments on influence relationships among attributes given by M
1
.
A1A2A3A4A5A6A7A8A9A10 A11 A12
A1 0 MW 0 MW 0 0 0 0 0 0 0 0
A2VW0 0 MS0 0 0 00 0 0 0
A3 VWM 0 VW0 VS VW0 MW0 0 0
A4 W S 0 0 W VW VW 0 VW 0 0 0
A500VWVW00000000
A6VWW0MWVW0 000 00 0
A70000VW0000000
A80000W0000000
A9WW0 S MS0 0 00 0 0 0
A10 MS VW W W W MW W W MW 0 0 0
A11 MW VW VW W W VW W 0 VW 0 0 0
A12 VW VW VW VW 0 W S S VS MS S 0
Table 7
The evaluation results with respect to A61 from the first ten respondents.
MO A I RQ
1 0.28 0.42 0.18 0.12 0 0
2 0 0.8 0.2 0 0 0
3 0.07 0.63 0.27 0.03 0 0
4 0 0.55 0.45 0 0 0
5 0.15 0.85 0 0 0 0
6 0.16 0.64 0.16 0.04 0 0
7 0.25 0.25 0.25 0.25 0 0
8 0.45 0.55 0 0 0 0
9 0.48 0.42 0.08 0.12 0 0
10 0 0.65 0.35 0 0 0
Table 8
The judgments of customer 1(M
1
) and the final average judgments on performance
and original importance for all attributes.
Performance
(C1)
Importance
(C1)
Performance
(average)
Importance
(average)
A1 [0.6, 0.9] [0.4, 0.5] [0.58, 0.85] [0.43, 0.53]
A2 [0.6, 0.9] [0.9, 1] [0.46, 0.67] [0.81, 0.97]
A3 [0.3, 0.4] [0.4, 0.5] [0.36, 0.46] [0.44, 0.54]
A4 [0.6, 0.9] [0.5, 0.6] [0.86, 0.99] [0.56, 0.62]
A5 [0.4, 0.5] [0.4, 0.5] [0.42, 0.54] [0.39, 0.49]
A6 [0.4, 0.5] [0.6, 0.9] [0.32, 0.42] [0.57, 0.8]
A7 [0.9, 1] [0.6, 0.9] [0.8, 0.9] [0.58, 0.84]
A8 [0.1, 0.3] [0.5, 0.6] [0.26, 0.38] [0.55, 0.74]
A9 [0.1, 0.3] [0.3, 0.4] [0.19, 0.36] [0.32, 0.42]
A10 [0.4, 0.5] [0.3, 0.4] [0.83, 0.93] [0.31, 0.41]
A11 [0.5, 0.6] [0.4, 0.5] [0.43, 0.53] [0.43, 0.53]
A12 [0.3, 0.4] [0.5, 0.6] [0.35, 0.45] [0.49, 0.59]
Table 9
The data on modifying attributes importance.
Kano
quality
category
p(PPI) Adjustment
model of I
Similarity
between
Pand I
IModified I
A1M1.000 I(1 (1 S))
2
0.743 0.472 0.258
A2O0.000 I(1 + (1 S)) 0.675 0.964 1.272
A3A0.100 I(1 + (1 S))
1/
2
0.920 0.489 0.508
A4O1.000 I(1 (1 S)) 0.661 0.590 0.389
A5O0.682 I(1 (1 S) 0.957 0.433 0.416
A6O0.000 I(1 + (1 S)) 0.677 0.743 0.981
A7O0.878 I(1 (1 S)) 0.842 0.783 0.658
A8O0.000 I(1 + (1 S)) 0.678 0.673 0.888
A9A0.159 I(1 + (1 S))
1/
2
0.901 0.350 0.367
A10 O1.000 I(1 (1 S)) 0.475 0.339 0.163
A11 O0.510 I(1 (1 S)) 0.998 0.476 0.471
A12 O0.000 I(1 + (1 S)) 0.864 0.544 0.616
1498 X. Geng, X. Chu / Expert Systems with Applications 39 (2012) 1492–1502
By using mðPÞ
0
15
mðNÞ
15
,a55 fuzzy relation matrix S
1
is ob-
tained for the Kano two-dimensional attribute classification:
S
1
¼
0000:18 0:42
0000:12 0:28
000 0 0
000 0 0
000 0 0
2
6
6
6
6
6
6
6
6
6
6
4
3
7
7
7
7
7
7
7
7
7
7
5
:
The two-dimensional attribute classification with respect to A6
is obtained based on Kano evaluation table (Table 2). The evalua-
tion results with respect to A6 considering the different attribute
classes from the first ten respondents are given in Table 7. Assume
that the threshold is selected as
a
= 0.5. When a measure in Table 7
is larger than or equal to
a
, ‘‘1’’ is used to replace this measure.
When a measure is smaller than
a
, ‘‘0’’ is used to replace this mea-
sure. Then the numbers of ‘‘1’’ appearing in different Kano attribute
classes are counted. The numbers of ‘‘1’’ appearing in different
attribute classes for A6 from the 50 respondents’ evaluations are
counted as: F
a
¼0:5
fM;O;A;I;R;Qg¼f14;34;2;0;0;0g.
Table 12
The vague total-relation matrix.
A1A2A3A4A5A6
A1 [0.003, 0.009] [0.086, 0.115] [0, 0.0001] [0.057, 0.093] [0.001, 0.006] [0, 0.002]
A2 [0.007, 0.038] [0.017, 0.034] [0, 0.0003] [0.146, 0.200] [0.003, 0.013] [0, 0.004]
A3 [0.004, 0.039] [0.086, 0.125] [0, 0.0004] [0.024, 0.070] [0.003, 0.018] [0.120, 0.142]
A4 [0.051, 0.087] [0.119, 0.166] [0, 0.0014] [0.019, 0.046] [0.021, 0.068] [0, 0.021]
A5 [0.001, 0.004] [0.001, 0.009] [0, 0.021] [0.009, 0.043] [0.0002, 0.003] [0, 0.004]
A6 [0.014, 0.046] [0.065, 0.102] [0, 0.0006] [0.073, 0.104] [0.002, 0.027] [0, 0.002]
A7 [0, 0.0002] [0, 0.0005] [0, 0.001] [0.0001, 0.002] [0.010, 0.050] [0, 0.0002]
A8 [0, 0.0003] [0, 0.0007] [0, 0.002] [0.0004, 0.003] [0.040, 0.071] [0, 0.0003]
A9 [0.046, 0.092] [0.032, 0.087] [0, 0.003] [0.123, 0.190] [0.093, 0.152] [0, 0.004]
A10 [0.075, 0.116] [0.030, 0.092] [0.049, 0.079] [0.035, 0.104] [0.027, 0.089] [0.066, 0.093]
A11 [0.063, 0.092] [0.011, 0.051] [0.000, 0.023] [0.055, 0.106] [0.052, 0.094] [0, 0.025]
A12 [0.033, 0.094] [0.029, 0.111] [0.014, 0.055] [0.033, 0.110] [0.029, 0.083] [0.019, 0.077]
A7A8A9A10 A11 A12
A1 [0.0006, 0.003] 0 [0, 0.002] 0 0 0
A2 [0.002, 0.006] 0 [0, 0.004] 0 0 0
A3 [0.0002, 0.022] 0 [0.030, 0.071] 0 0 0
A4 [0.010, 0.031] 0 [0, 0.021] 0 0 0
A5 [0.0001, 0.002] 0 [0, 0.002] 0 0 0
A6 [0.0007, 0.003] 0 [0, 0.002] 0 0 0
A7 [0, 0.0001] 0 [0, 0.0001] 0 0 0
A8 [0, 0.0001] 0 [0, 0.0002] 0 0 0
A9 [0.001, 0.0068] 0 [0, 0.0004] 0 0 0
A10 [0.060, 0.095] [0.010, 0.040] [0.062, 0.098] 0 0 0
A11 [0.051, 0.084] 0 [0, 0.024] 0 0 0
A12 [0.133, 0.202] [0.131, 0.217] [0.158, 0.195] [0.130, 0.180] [0.100, 0.143] 0
Table 11
The aggregated normalized direct-relation matrix.
A1A2A3A4A5A6
A1 0 [0.079, 0.101] 0 [0.045, 0.074] 0 0
A2 [0.000, 0.021] 0 0 [0.142, 0.191] 0 0
A3 [0.000, 0.021] [0.076, 0.096] 0 [0.000, 0.021] 0 [0.119, 0.141]
A4 [0.049, 0.078] [0.113, 0.148] 0 0 [0.021, 0.063] [0.000, 0.021]
A5 0 0 [0, 0.021] [0.009, 0.039] 0 0
A6 [0.006, 0.035] [0.056, 0.082] 0 [0.063, 0.083] [0, 0.021] 0
A70000[0.013, 0.047] 0
A80000[0.038, 0.071] 0
A9 [0.045, 0.075] [0.014, 0.049] 0 [0.116, 0.163] [0.093, 0.135] 0
A10 [0.072, 0.094] [0.012, 0.045] [0.049, 0.077] [0.016, 0.052] [0.018, 0.056] [0.064, 0.084]
A11 [0.056, 0.078] [0.000, 0.021] [0.000, 0.021] [0.050, 0.082] [0.054, 0.080] [0.000, 0.020]
A12 [0.008, 0.036] [0.017, 0.054] [0.009, 0.036] [0.000, 0.021] 0 [0.014, 0.049]
A7A8A9A10 A11 A12
A1000000
A2000000
A3 [0.000, 0.021] 0 [0.034, 0.070] 0 0 0
A4 [0.006, 0.033] 0 [0.000, 0.021] 0 0 0
A5000000
A6000000
A7000000
A8000000
A9000000
A10 [0.057, 0.087] [0.009, 0.039] [0.065, 0.085] 0 0 0
A11 [0.052, 0.079] 0 [0.000, 0.020] 0 0 0
A12 [0.116, 0.165] [0.131, 0.208] [0.151, 0.172] [0.129, 0.179] [0.098, 0.143] 0
X. Geng, X. Chu / Expert Systems with Applications 39 (2012) 1492–1502 1499
Therefore, A6 is a one-dimensional attribute (O). The results of
the analysis of FKQ and FKM are: A1 is a must-be attribute (M);
A2, A4, A5, A6, A7, A8, A10, A11 and A12 are one-dimensional attri-
butes (O); and A3 and A9 are attractive attributes (A).
5.2. Obtaining the final importance of attributes
Each customer gives out his judgments on performance and ori-
ginal importance for all attributes elicited. Table 8 presents the
judgments of customer 1(C
1
) and the final average judgments of
all customers:
Step 1. Revising the original importance of each attribute by the
adjustment models based on Kano’s model.
According to the Eq. (1), the possibility of PPIfor each attri-
bute can be calculated .They are 1, 0, 0.1, 1, 0.68, 0, 0.88, 0, 0.16,
1, 0.51, 0, respectively. According to the Kano quality category of
each attribute, the adjustment model of original importance for
each attribute can be determined as shown in Table 9. The modi-
fied importance of each attribute can be obtained after calculating
the similarity between attribute performance and attribute origi-
nal importance by Eq. (2). The attributes’ original importance and
the modified importance are given out in Table 9.
Step 2. Obtaining the final importance of attributes based on
DEMATEL.
Survey questionnaires are sent to 15 experienced experts and
customers to obtain the direct influence relationships among the
12 attributes. The judgments of M
1
are shown in Table 10, and
the aggregated normalized direct-relation matrix is depicted in Ta-
ble 11. The vague total-relation matrix can be obtained according
to Eqs. (13) and (14) as shown in Table 12. The sum of every row
and every column can be transformed into crisp value by Eqs.
(15) and (16), and then D+R and DRof every attribute can be
calculated to reflect the prominence and the net influence level
of the attribute, respectively. The results of D+R and DRare
shown in Table 13. The DEMATEL causal diagram can be drawn
according to D+Rand DRof each attribute, as shown in Fig. 3.
As shown in the causal diagram, the PSS quality attributes are visu-
ally divided into the cause group including A3, A9, A10, A11, and
A12, and the effect group which is composed of attribute A1, A2,
A4, A5, A6, A7, A8, and A9. The causal diagram can provide valuable
information for making decisions. On one hand, the net influence
level of every attribute can be used to revise the importance of
the attribute, and IPA can be conducted taking into account the
mutual influence relationships among all attributes; On the other
hand, qualitative decision cues can be identified for supporting
making strategies. If Company H wishes to reach a high level of
performance in terms of the effect attributes group, it must control
and pay much attention to the cause attributes group.
Considering the mutual influence relationships among attri-
butes, DRvalue of each attribute is added to the modified impor-
tance of each attribute in Table 8. The final importance of all
attributes is 0.01, 0.823, 0.761, 0.104, 0.011, 0.911, 0.334,
0.747, 0.448, 0.618, 0.714, and 1.755, respectively. The attribute
whose final importance is negative can be regarded as much less
important for company’s competitiveness, and it is not considered
in IPA map. The revised IPA map is shown in Fig. 4.A1, A4 and A5
are not considered in the revised IPA. In order to do a contrastive
analysis, Original IPA map can be drawn according to the perfor-
mance and original importance of all attributes, as shown in
Fig. 3. The DEMATEL causal diagram.
Fig. 4. The revised IPA map.
Table 13
The data on D+Rand DRof each attribute.
Din interval
number
Din
crisp
number
Rin interval
number
Rin
crisp
number
D+R DR
A1 [0.148, 0.231] 0.189 [0.296, 0.618] 0.457 0.646 0.268
A2 [0.175, 0.299] 0.237 [0.477, 0.896] 0.686 0.923 0.449
A3 [0.267, 0.488] 0.378 [0.063, 0.186] 0.125 0.503 0.253
A4 [0.220, 0.442] 0.331 [0.575, 1.072] 0.824 1.155 0.493
A5 [0.011, 0.089] 0.050 [0.279, 0.674] 0.477 0.527 0.427
A6 [0.154, 0.287] 0.220 [0.205, 0.375] 0.290 0.510 0.070
A7 [0.010, 0.055] 0.032 [0.259, 0.453] 0.356 0.388 0.324
A8 [0.040, 0.076] 0.058 [0.141, 0.257] 0.199 0.257 0.141
A9 [0.295, 0.539] 0.417 [0.249, 0.423] 0.336 0.753 0.081
A10 [0.414, 0.806] 0.610 [0.130, 0.180] 0.155 0.765 0.455
A11 [0.231, 0.499] 0.365 [0.100, 0.143] 0.122 0.487 0.243
A12 [0.811, 1.47] 1.139 0 0 1.139 1.139
1500 X. Geng, X. Chu / Expert Systems with Applications 39 (2012) 1492–1502
Fig. 5. Compared with Fig. 5,A7 is relocated from Quadrant I to
Quadrant IV, and it shows that A7 is affected a lot by other quality
attributes and its importance is lowered significantly. Therefore,
improvement should not be done directly upon A7. A8 is relocated
from Quadrant II to Quadrant III. Its importance is also lowered be-
cause it receives much impact from other quality attributes, and
the urgency of improvement is decreased. A12 is relocated from
Quadrant III to Quadrant II. It shows that its importance increases
a lot. From the Fig. 3, it can be seen that A12 affects many quality
attributes, but not receives little affection from other attribute.
Therefore, improvement should be carried out directly upon A12.
6. Conclusions
Identifying improvement opportunities to make improving de-
sign strategies plays a critical role in providing satisfying PSSs to
customers. In order to implement customer satisfaction evaluation
for supporting PSS design, a new IPA is proposed by absorbing
advantages and overcoming shortcomings of the former re-
searches. The major characteristics of this research are summa-
rized as follows:
(1) Vague set theory is employed in IPA to deal with the uncer-
tainty and vagueness in evaluation process due to its advan-
tage in supporting the opposing evidences over fuzzy set
theory.
(2) Kano’s model is incorporated into IPA to consider the nonlin-
ear relationships between attributes performance and over-
all customer satisfaction. The impact of attribute
performance on attribute importance is expressed in adjust-
ing attribute original importance considering Kano’s model
and attribute performance level. Three adjustment models
based on vague sets are given out aiming at attributes
belonging to three different Kano quality categories: must-
be, one-dimension, and attractive attribute.
(3) DEMATEL based on vague sets is used to analyze the mutual
influence relationships among attributes, and the net influ-
ence level of each attribute on other attributes is used to
revise attribute importance to consider the impact of these
mutual relationships on strategies making.
Acknowledgements
The project was supported by National Natural Science Founda-
tion, China (No. 51075261), Shanghai Science and Technology
Innovation Action Plan (No. 09dz1124600, No. 10dz1121600),
Shanghai Jiao Tong University Innovation Fund For Postgraduates.
The authors would also like to express their grateful appreciation
to the anonymous referees for their helpful comments to improve
the quality of this paper.
References
Abalo, J., Varela, J., & Manzano, V. (2007). Importance values for importance–
performance analysis: A formula for spreading out values derived from
preference rankings. Journal of Business Research, 60(2), 115–121.
Aurich, J. C., Fuchs, C., & DeVries, M. F. (2004). An approach to life cycle oriented
technical service design. CIRP Annals – Manufacturing Technology, 53(1),
151–154.
Aurich, J. C., Fuchs, C., & Wagenknecht, C. (2006). Life cycle oriented design of
technical product-service systems. Journal of Cleaner Production, 14(17),
1480–1494.
Boran, F. E., Gen, S., Kurt, M., & Akay, D. (2009). A multi-criteria intuitionistic fuzzy
group decision making for supplier selection with TOPSIS method. Expert
Systems with Applications, 36(8), 11363–11368.
Chen, J.-K., & Chen, I. S. (2010). Using a novel conjunctive MCDM approach based on
DEMATEL, fuzzy ANP, and TOPSIS as an innovation support system for
Taiwanese higher education. Expert Systems with Applications, 37(3),
1981–1990.
Chen, Y. -C., Lien, H. -P., Tzeng, G. -H., & Yang, L.S. (2009). Combined DEMATEL
technique with VIKOR method for improving environment-watershed plan
strategy. In EURO conference in Bonn 2009, July 5–8 (p. TA-37).
Chen, Y.-C., Lien, H.-P., & Tzeng, G.-H. (2010). Measures and evaluation for
environment watershed plans using a novel hybrid MCDM model. Expert
Systems with Applications, 37(2), 926–938.
Chu, R. K. S., & Choi, T. (2000). An importance–performance analysis of hotel
selection factors in the Hong Kong hotel industry: A comparison of business and
leisure travellers. Tourism Management, 21(4), 363–377.
Deng, W. J. (2007). Using a revised importance–performance analysis approach: The
case of Taiwanese hot springs tourism. Tourism Management, 28(5), 1274–1284.
Deng, W.-J., Chen, W.-C., & Pei, W. (2008). Back-propagation neural network based
importance–performance analysis for determining critical service attributes.
Expert Systems with Applications, 34(2), 1115–1125.
Enright, M. J., & Newton, J. (2004). Tourism destination competitiveness: A
quantitative approach. Tourism Management, 25(6), 777–788.
Gabus, A., & Fontela, E. (1973). Perceptions of the world problematique:
Communication procedure, communicating with those bearing collective
responsibility (DEMATEL Report No. 1). Geneva, Switzerland: Battelle Geneva
Research Centre.
Gabus, A., & Fontela, E. (1972). World problems an invitation to further thought within
the framework of DEMATEL. Geneva, Switzerland: Battelle Geneva Research
Centre.
Garver, M. S. (2002). Using data mining for customer satisfaction research.
Marketing Research, 14(1), 8–12.
Gau, W., & Buehrer, D. (1993). Vague sets. IEEE Transactions on Systems, Man and
Cybernetics, 23(2), 610–614.
Hansen, E., & Bush, R. J. (1999). Understanding customer quality requirements:
Model and application. Industrial Marketing Management, 28(2), 119–130.
Hu, H.-Y., Lee, Y.-C., & Yen, T.-M. (2009). Amend importance–performance analysis
method with Kano’s model and DEMATEL. Journal of Applied Sciences, 9(10),
1833–1846.
Hu, H.-Y., Lee, Y.-C., Yen, T.-M., & Tsai, C.-H. (2009). Using BPNN and DEMATEL to
modify importance–performance analysis model – A study of the computer
industry. Expert Systems with Applications, 36(6), 9969–9979.
Huan, T.-C., Beaman, J., & Shelby, L. B. (2002). Using action-grids in tourism
management. Tourism Management, 23(3), 255–264.
Huang, C.-Y., Shyu, J. Z., & Tzeng, G. H. (2007). Reconfiguring the innovation policy
portfolios for Taiwan’s SIP Mall industry. Technovation, 27(12), 744–765.
Kano, N., Seraku, N., Takahashi, F., & Tsuji, S. (1984). Attractive quality and must-be
quality. Quality, 14(2), 39–48.
Kim, B. G., Park, S. C., & Lee, K. J. (2007). A structural equation modeling of the
Internet acceptance in Korea. Electronic Commerce Research and Applications,
6(4), 425–432.
Lee, Y.-C., & Huang, S.-Y. (2009). A new fuzzy concept approach for Kano’s model.
Expert Systems with Applications, 36(3), 4479–4484.
Lee, Y.-C., Li, M.-L., Yen, T.-M., & Huang, T.-H. (2010). Analysis of adopting an
integrated decision making trial and evaluation laboratory on a technology
acceptance model. Expert Systems with Applications, 37(2), 1745–1754.
Lin, C.-J., & Wu, W.-W. (2008). A causal analytical method for group decision-
making under fuzzy environment. Expert Systems with Applications, 34(1),
205–213.
Lo, C., Wang, P., & Chao, K. (2006). A fuzzy group-preferences analysis method for
new-product development. Expert Systems with Applications, 31(4), 826–834.
Manzini, E., & Vezzoli, C. (2003). A strategic design approach to develop sustainable
product service systems: Examples taken from the ‘environmentally friendly
innovation’ Italian prize. Journal of Cleaner Production, 11(8), 851–857.
Martilla, J., & James, J. (1977). Importance–performance analysis. The Journal of
Marketing, 41(1), 77–79.
Fig. 5. The original IPA map with attributes performance and original importance.
X. Geng, X. Chu / Expert Systems with Applications 39 (2012) 1492–1502 1501
Matzler, K., Bailom, F., Hinterhuber, H. H., Renzl, B., & Pichler, J. (2004). The
asymmetric relationship between attribute-level performance and overall
customer satisfaction: A reconsideration of the importance–performance
analysis. Industrial Marketing Management, 33(4), 271–277.
Matzler, K., & Hinterhuber, H. (1998). How to make product development projects
more successful by integrating Kano’s model of customer satisfaction into
quality function deployment. Technovation, 18(1), 25–38.
Matzler, K., & Sauerwein, E. (2002). The factor structure of customer satisfaction.
International Journal of Service Industry Management, 13(4), 314–332.
Matzler, K., Sauerwein, E., & Heischmidt, K. A. (2003). Importance–performance
analysis revisited: The role of the factor structure of customer satisfaction.
Service Industries Journal, 23(2), 112–129.
Mont, O. K. (2002). Clarifying the concept of product-service system. Journal of
Cleaner Production, 10(3), 237–245.
Morelli, N. (2006). Developing new product service system (PSS): Methodologies
and operational tools. Journal of Cleaner Production, 14(17), 1495–1501.
Ou Yang, Y. P., Shieh, H. M., Leu, J. D., & Tzeng, G. H. (2008). A novel hybrid MCDM
model combined with DEMATEL and ANP with application. International Journal
of Operations Research, 5(3), 160–168.
Ryan, C., & Huyton, J. (2002). Tourists and Aboriginal people. Annals of Tourism
Research, 29(3), 631–647.
Sakao, T., & Shimomura, Y. (2007). Service engineering: A novel engineering
discipline for producers to increase value combining service and product.
Journal of Cleaner Production, 15(6), 590–604.
Sakao, T., Shimomura, Y., Sundin, E., & Comstock, M. (2009). Modeling design
objects in CAD system for service/product engineering. Computer-Aided Design,
41(3), 197–213.
Seyed-Hosseini, S. M., Safaei, N., & Asgharpour, M. J. (2006). Reprioritization of
failures in a system failure mode and effects analysis by decision making trial
and evaluation laboratory technique. Reliability Engineering & System Safety,
91(8), 872–881.
Tonge, J., & Moore, S. A. (2007). Importance-satisfaction analysis for marine-park
hinterlands: A Western Australian case study. Tourism Management, 28(3),
768–776.
Tzeng, G.-H., Chiang, C.-H., & Li, C.-W. (2007). Evaluating intertwined effects in e-
learning programs: A novel hybrid MCDM model based on factor analysis and
DEMATEL. Expert Systems with Applications, 32(4), 1028–1044.
Williams, A. (2007). Product service systems in the automobile industry:
Contribution to system innovation? Journal of Cleaner Production, 15(11–12),
1093–1103.
Wu, W.-W. (2008). Choosing knowledge management strategies by using a
combined ANP and DEMATEL approach. Expert Systems with Applications,
35(3), 828–835.
Wu, W.-W., & Lee, Y.-T. (2007). Developing global managers’ competencies
using the fuzzy DEMATEL method. Expert Systems with Applications, 32(2),
499–507.
Ye, J. (2007). Improved method of multi criteria fuzzy decision-making based on
vague sets. Computer-Aided Design, 39(2), 164–169.
Zhang, H. Q., & Chow, I. (2004). Application of importance–performance model in
tour guides’ performance: Evidence from mainland Chinese outbound visitors
in Hong Kong. Tourism Management, 25(1), 81–91.
Zhang, D., Huang, S., & Li, F. (2004). Approach to measuring the similarity
between Vague sets. Huazhong Keji Daxue Xuebao (Ziran Kexue Ban)/Journal of
Huazhong University of Science and Technology (Natural Science Edition), 32(5),
59–60.
Zhang, D., Zhang, J., Lai, K.-K., & Lu, Y. (2009). An novel approach to supplier
selection based on vague sets group decision. Expert Systems with Applications,
36(5), 9557–9563.
1502 X. Geng, X. Chu / Expert Systems with Applications 39 (2012) 1492–1502