Department of Multimedia and Entertainment Science, Southern Taiwan University

advertisement
A General Framework for Kansei Engineering System
Chih-Chieh Yang and Meng-Dar Shieh*
Department of Multimedia and Entertainment Science, Southern Taiwan University
No. 1, Nantai Street, Yongkang City, Tainan County, Taiwan 71005
Department of Industrial Design, National Cheng Kung University, Tainan, Taiwan
70101*
Abstract
In this study, a general framework for Kansei engineering system (KES) is
proposed. KES, as a kind of consumer-oriented technology, facilitates product designs
that elicit positive consumer affective responses (CARs) and investigates the
influences of the product form features (PFFs). This proposed KES framework
extends those already existing by introducing three important concepts: 1) consumer
segmentation (CS), 2) affective response dimension selection (ARDS), and 3) product
form feature selection (PFFS). The purpose of the ARDS approach is to choose
suitable adjectives to describe CARs. The PFFS methodology aims to pin point
critical PFFs that influence the CARs of the product design. The proposed framework
for KES brings out several useful instruments for the application in product design
and demonstrates its potential for facilitating the product development process. This
paper describes important concepts related to this proposed framework and reviews
from the literature a suitable methodology, which can be used to implement the
procedures of this framework. Preliminary results from the authors’ recent studies are
also summarized. However, more future studies using different kinds of products are
still needed to verify the effectiveness of the proposed KES framework.
Keywords: Affective responses; Product form features; Consumer segmentation;
Affective response dimension selection; Product form feature selection.
1. Introduction
The way a product looks is one of the most important factors affecting a
consumer’s purchasing decision. The task of the product designer is to manipulate the
product form features (PFFs) to produce specific styles that satisfy the consumer’s
expectations. Of course, how to do this is of great importance. Traditionally, the
success of a product’s design depended on the designers’ artistic sensibilities, which
1
quite often did not meet with great acceptance in the marketplace. Consequently,
researchers have conducted many product design studies to get a better insight into
consumers’ subjective perceptions. The most notable research is the Kansei
engineering system (KES). As shown in Fig. 1, the basic assumption of KES is the
cause-and-effect relationship between the PFFs and CARs [1]. KES was developed in
Japan in the 1970’s and not revealed to the product design community until papers
published by Mitsuo Nagamachi and his colleges started to appear in journals [2, 3].
In the last decade, KES research has been extensively carried out in Asia. In Taiwan,
the most notable researches are the works of Shih-Wen Hsiao [4-6] from National
Cheng Kung University. In Korea, the user interface laboratory of Pohang University
of Science and Technology, which is lead by Sung H. Han, has conducted extensive
research applying the KES approach to product design [1, 7, 8]. However, in our
experience, some misunderstandings regarding KES research still exist in the product
design community. A product designer may question the usefulness of KES. Actually,
the purpose of KES is intended neither to replace the designer’s creativity and
aesthetics nor to generate product alternatives using computers without any human
supervision. Rather, KES offers the opportunity to systematically process information
on large amounts of product samples and analyze the corresponding CARs. KES
researchers still encounter many fundamental problems and are not equipped with a
suitable methodology to solve them, thus delaying the development of real-world
applications for product designs.
As a typical KES process, Nagamachi [9] suggested the following procedures: (1)
select the object; (2) collect adjectives; (3) understand the structural meanings of the
adjectives; (4) prepare slides or samples of the materials; (5) evaluate emotions; (6)
do statistical analysis; and (7) build an expert system. After a decade, Schutte et al.
[10] proposed a general framework based on a similar rationale to that of
Nagamachi’s [9]. Their framework consists of the following steps: (1) choose the
product domain; (2) span the semantic space; (3) span the product properties space;
and (4) synthesize. The core of KES is to construct a prediction model by using the
PFFs as input and the CARs as predictive output (see Fig. 1). The “do statistical
analysis” step in [9] and the “synthesize” step in [10] belong to this aspect. In addition,
the construction of the prediction model can be regarded as a classification or
regression problem. In this study, a general framework for KES is proposed which
extends those of Nagamachi’s [9] and Schutte et al.’s [10]. Three important concepts:
1) consumer segmentation (CS), 2) affective response dimension selection (ARDS),
and 3) product form feature selection (PFFS), are introduced to extend the existing
KES framework. The concept of CS was borrowed from the field of market research
to aid in reassembling consumers with heterogeneous opinions into homogenous
2
groups with similar preferences. Second, the purpose of the ARDS approach is to
choose suitable adjectives to describe CARs. Finally, the PFFS methodology aims to
pin point critical PFFs that influence the CARs of the product design. This proposed
framework for KES introduces several useful applications in product design and
demonstrates its potential to facilitate the product development process. The
remainder of this paper is organized as follows: Section 2 reviews important literature
and the suitable methodology related to the proposed framework. Section 3 presents
the proposed KES framework and its implementation procedures. Section 4 suggests
some future directions within the proposed framework. Section 5 gives some brief
concluding remarks.
< Insert Fig. 1 about here >
2. Background review
2.1. Description of affective responses with adjectives
There are some differences in the affective responses of professional product
designers and consumers [11]. Different kinds of adjectives must be used to describe
their affective responses. In most research, this problem is often neglected. The
framework proposed by Lamb and Kallal [12] can be adapted to elicit two types of
adjectives to describe affective responses: aesthetic adjectives for product designers
and expressive adjectives for consumers
Aesthetic adjectives are more suitable for designers to describe the affective
responses of the product form features (PFFs) since the role of product designers is to
manipulate form elements to make them aesthetically appealing. These kinds of
adjectives are closely related to aesthetic or style principles. Most aesthetic adjectives
can be interpreted using certain aesthetic principles such as harmony, balance,
uniformity, etc. These adjectives are often closely related to a product’s attributes such
as color, material and texture. Chen [13] has successfully combined these kinds of
adjectives (Table 1) with computerized PFFs.
Due to their properties of high diversity and lack of easy interpretation,
expressive adjectives are more suitable for consumers to express their sensations
toward product samples. These kind of adjectives can be found in the research of [4,
11, 14]. Some examples of expressive adjectives are shown in Table 2. Notice that
when consumers are asked to express their reactions using aesthetic adjectives, their
answers are ambiguous and not very satisfying. Expressive adjectives also provide us
with an excellent tool for studying consumer preferences and determining new
3
product positioning.
< Insert Table 1 about here >
< Insert Table 2 about here >
2.2. Product form feature representation
PFFs in KES are represented by three kinds of schemes in the authors’ survey: a
mainly morphological analysis based method, an object-oriented method, and, a
level-of-detail method. The morphological analysis based method [15] decomposes a
product into several components. For example, an office chair might be decomposed
into its parts such as back, seat, back support, armrest, base, etc. Then, an examination
would be made of every possible choices for each component (see Fig. 2) [6]. In this
way, PFFs can be represented in a simple and intuitive manner. Numerical and
parametric representations are also easy to obtain. In cases where the product samples
are not quite dissimilar, this scheme is very suitable for representing PFFs. However,
different methods are possible when decomposing different kinds of products. It is
extremely difficult to find a unified method to represent all product samples.
Object-oriented representation overcomes the limitations of the morphological
analysis based method by using the concept of class-subclass, which is capable of
representing more complicated product samples with a greater deviation in product
components [16]. All attributes in the parent class will be inherited by the children’s
class, while different attributes of their own will be listed in the child class. This
representation is extremely useful when the PFFs of different product samples lack a
unique, natural identification. All the product form component have the same
attributes such as size, color, and material and can be defined in the parent class, but
also have their own unique attributes that can be defined in the child class (see Fig. 3).
These characteristics make the object-oriented representation more flexible in
representing PFFs than the morphological analysis based representation.
The level-of-detail representation also attempts to use a unified framework to
describe product form [13, 17]. It is based on the fact that, in the stage of conceptual
design, level-of-detail concepts exist naturally when product designers want to
express a product form design. They tend to rough sketch form elements first, and
then consider the unifying relationships of these elements, finally adding the details
such as color, material, textures, etc. In this unified framework, different products can
be represented regardless of types. Research by Wallace and Jakiela [18] proposed a
similar concept and applied it to the design of consumer electronics. They used
4
pre-defined components to set up the initial organization, and then constructed a
smooth outer surface to produce better product appearance (see Fig. 4).
< Insert Fig. 2 about here >
< Insert Fig. 3 about here >
< Insert Fig. 4 about here >
2.3. Nonlinear prediction model of affective responses
One of the frequently encountered problems when modeling CARs in KES is
how to deal with the nonlinear relationship between the PFFs and the CARs. Various
methods can be used to construct a prediction model for CARs. The most used
techniques in the product design field such as multiple linear regression (MLR) [8],
quantification theory type I (QT1) [19], and partial least squares regression (PLSR)
[20] are all heavily dependent on an assumption of linearity and can not deal
effectively with a nonlinear relationship. In addition, prior to establishing the
prediction model, data simplification or variable screening is often needed to obtain
better results. Fuzzy regression analysis [2] and other methods suffer from the same
shortcomings [7].
To deal with the nonlinearity of many-to-many mapping between variables, the
neural network (NN) is a good candidate for building such a prediction model. A few
researches have illustrated the use of NN in the product design field. Hsiao and Huang
[21] demonstrated the ability of NN to deal with the nonlinear relationships between
the PFFs. In later research by Hsiao and Tsai [4], NN was used as part of a hybrid
framework for a product form search. However, NN suffers from a number of
shortcomings. NN is considered a “black-box” necessitating numerous control
parameters and it is difficult to obtain a stable solution. Vapnik [22] developed a new
kind of NN algorithm called support vector machine (SVM). SVM has been shown to
provide better performance than traditional learning techniques [23]. SVM is also
known for its elegance in solving nonlinear problems using the “kernels” technique,
which automatically carries out a nonlinear mapping to a feature space. With the
introduction of appropriate loss function, SVM can be extended to solve function
estimation problems. This is known as support vector regression (SVR). Despite
being endowed with a number of attractive properties, SVM/SVR has yet to be
applied widely in the field of product design.
5
2.4. Consumer segmentation
Since Smith [24] presented the concept of consumer segmentation (CS), it has
become an important marketing strategy. It is known that the needs of consumers are
often diverse and heterogeneous. By distinguishing different homogeneous groups, a
more precise adjustment of product alternatives to suit these consumer groups can be
made. In order to recognize consumer heterogeneity effectively, choosing the relevant
segmentation variables and the methods used to construct the segments are two
critical factors. Different consumer characteristics such as demographics, lifestyle,
socio-economic factors, purchasing behavior, attitudes and preferences toward
product alternatives, etc., can be used as segmentation variables to construct segments.
Bock and Uncles [25] identified five distinct types of CS according to these
segmentation variables. Since the CAR is supposed to measure consumers’ needs, the
segmentation variables should be capable of reflecting their preferences toward a
product’s design. The variables such as demographics, lifestyle and other factors,
which are frequently used in market research, do not directly relate to the product
itself and are therefore not suitable for CS in product design. Rather, the preference
data of consumers evaluated on the representative product samples are more suitable
for use as segmentation variables for CAR research in product design.
When constructing CS, cluster analysis is commonly applied. The purpose of
cluster analysis is to construct homogenous groups with respect to the segmentation
variables. Various methods have been used for CS including K-means clustering,
self-organizing maps, fuzzy clustering, and support vector clustering. Of these
methods, fuzzy clustering proves to be the most robust method for CS when compared
to other non-fuzzy clustering [26]. Fuzzy clustering is capable of dealing with the
overlapping clusters and is stable even in the presence of outliers. The main reason for
its superior performance is the production of the fuzzy partition in which each data
point is associated with a set of membership grades indicating the degrees to which
this point belongs in the different clusters. The distances between data points in the
clustering results also provide useful information to determine their relative
importance. Consequently, fuzzy clustering is very suitable for CS application.
2.5. Selection of representative affective response dimensions
In order to obtain CARs, the semantic differential (SD) experiment [27] is often
conducted. This asks consumers to evaluate product samples using chosen adjectives.
Factor analysis (FA) is the most frequently adopted technique for analyzing the
evaluated scores obtained from SD experiments [28]. FA has been used to study the
6
differences in product form perception between consumers and designers [11]. In a
recent study by Hsiao and Chen [29], FA was used to examine the fundamental
dimensions of affective responses to different kinds of products of various sizes. In
the KES literature, many efforts have been made to study the CARs using FA, such as
those described in [14, 30-32]. In fact, FA belongs to the feature extraction method,
which reduces input dimensions into fewer latent dimensions. By applying such a
feature extraction technique, similar adjectives can be merged into factors according
to the CARs. Product designers can then examine and interpret the effects of the
adjectives by analyzing the factor loadings of the original adjectives on the latent
factors.
Currently, to the best of the authors’ knowledge, there is no KES research on
extracting representative affective response dimensions in an effective manner.
Similar to the ARDS problem discussed in this study, the field of sensory analysis also
addresses the need to select representative evaluation scales for measuring the
subjective perceptions of consumers. Sahmer and Qannari [33] offers several
strategies for selecting a subset of sensory attributes. The simplest solution is simply
to choose the attributes with large factor loading (in absolute value) of the latent
factors. The second strategy provides an alternative by applying clustering techniques
to the factor loadings, such as the method for the clustering of variables (COV)
proposed by Vigneau and Qannari [34]. The attributes can be arranged into
homogenous clusters according to their factor loadings. The third strategy makes use
of Procrustes analysis (PA) to select critical attributes while preserving as much as
possible the multivariate structure of the data. The PA approach for selecting a subset
of variables was first introduced by Krzanowski [35], and was combined with
principal component analysis (PCA) to process sensory profiling data. The superiority
of this approach lies in it providing direct measurements of the discrepancy between
the selected subset and the full set.
2.6. Selection of critical form features
As was first mentioned in Han and Hong [1], an important KES issue is that the
problem of PFFS according to consumers’ perceptions has not been intensively
investigated. The subjective perceptions of consumers are often influenced by a wide
variety of form features. The number of form features could be many and might be
highly correlated to each other. Consequently, manual inspection of the relative
importance of the PFFs and picking out the most important consumer pleasing
features is a difficult task. In the product design field, critical design features are often
arrived at based on the opinions of experts or focus groups. However, the selection of
7
features based on expert opinion often lacks objectivity. Only a few attempts have
been made to overcome these shortcomings in the PFFS process. Han and Hong [1]
uses several traditional statistical methods for screening critical design features
including principal component regression (PCR), cluster analysis, and partial least
squares (PLS). In the study of Han et al. [8], a genetic algorithm-based PLS method is
applied to screen design variables. Therefore, as a research topic PFFS methodology
is of great value and is still waiting for improvement.
Although the existing literature on PFFS is sparse, the problem of feature
selection can be found in many other fields besides that of product design. The nub of
the problem is how to find the subset of PFFs with the least possible generalization
errors and to select the smallest possible subset with a high predictive capability.
Different approaches have been proposed for solving the feature selection problem
including rough sets [36], neuro-fuzzy [37], and SVM [38, 39]. Another crucial issue
in solving the PFFS problem is how to deal with the correlations between PFFs and
CARs [2, 7]. The superiority of applying nonlinear prediction models, such as NN or
SVM, instead of MLR, QT1 or other linear models has already been addressed in Sect.
2.3.
3. The proposed framework for Kansei engineering system
In this study, a seven-step general framework is proposed for the KES research
as shown in Fig. 5. The steps of this framework and their relationship to the
methodology of KES are described in this section. Several achievements adapted from
the authors’ current research, which is in the spirit of the proposed framework, are
also reported.
< Insert Fig. 5 about here >
3.1. Selection of representative products
At the beginning of a proposed framework, researchers collect as many products
as possible from the marketplace. Experienced product designers examine these
product samples and take note of their characteristics. A reduced set of representative
products is used in Step 2 for selecting representative adjectives while the whole set
of product samples can be used to construct the prediction model of affective
responses in Step 6.
3.2. Selection of representative adjectives
8
Often, various adjectives can be used to describe CARs. Thus, redundant or
similar adjectives must be screened out. The mental capacity of consumers is limited
and the number of adjectives used to conduct an SD experiment should not be too
many to assure reasonable mental consistency. Therefore, researchers should prepare
an initial set of adjectives collected from a pilot test and choose the most
representative ones from that. Since there are many initial adjectives, an SD
experiment is first conducted in order for the consumers to evaluate a limited amount
of representative product samples obtained from Step 1. The ARDS methodology is
then used to analyze the results of the SD experiment and extract the representative
adjectives. For example, in a recent study by Yang and Shieh [40], an approach
combining FA and PA was used to select representative affective dimensions for
mobile phone design from an initial set of 22 pairwise adjectives adapted from [11]. A
backward elimination process based on PA is capable of determining the relative
importance of adjectives in each step according to the calculated residual sum of
squared differences (RSSDs).
3.3. Segmentation of consumers
According to market research, the needs of consumers are often diverse and
heterogeneous. By distinguishing different homogenous groups, a more precise
adjustment of product alternatives to these consumer groups can be made The CAR
data can also be used as segmentation variables to construct segments of consumers.
Typically, the CAR scores of the consumers in the same segment for each product
samples are averaged before conducting further analysis.
3.4. Determination of product form features
The characteristics of a product are mostly based on its form features. They also
provide the foundation for consumers to distinguish different products. Different
categories of products may have different ways of decomposing their PFFs. These
PFFs are treated as an input vector when training a prediction model of CAR. Three
kinds of PFF representation are reviewed in Sect. 2.2. As a simple and intuitive
method, morphological analysis can be used to decompose the PFFs of product
samples. These PFFs of product samples also need to be encoded into a numerical
format. An example of PFFs for mobile phone design used in [41, 42] is listed in
Table 3.
9
< Insert Table 3 about here >
3.5. Experimental and questionnaire design
To understand the relationship between PFFs and CARs, subjective perception
data of consumers must be collected. This can be achieved by conducting a second
SD experiment in the form of questionnaires to obtain CAR data on selected product
samples. The presentation order of product samples also needs to be randomized to
avoid systematic effects. Moreover, the experimental results of evaluating product
samples based on too many adjectives in too long a period of time inevitably induces
more errors [2]. Thus, the ARDS method in Step 2 is very helpful at the questionnaire
design stage for reducing mental over loading of the subjects.
3.6. Constructing a prediction model of affective responses
According to the basic assumptions of KES shown in Fig. 1, two kinds of
prediction models can be constructed to interrelate PFFs and CARs: the classification
based model and the regression based model. The classification based prediction
model aims to correctly discriminate different CARs according to the input PFFs.
Although more than one adjective can be used as a class label to formulate a
multiclass problem, the degree to which a product sample satisfies a specific adjective
can not be obtained from the predictive results. On the other hand, as a predictive
output, the regression based prediction model can only deal with one CAR at a time.
However, an exact CAR value of input product samples can be computed from a
regression based prediction model. To help product designers develop appealing
products in a effective manner, the authors propose an approach based on multiclass
fuzzy SVM [41]. This constructs a classification model of product form design using
five single adjectives as class labels (sports, simplicity, female, plain, and business).
The one-versus-one (OVO) method is adopted to handle the multiclass problem by
breaking it down into various two-class problems. The experimental results using
mobile phone designs demonstrated the superiority of the nonlinear SVM model,
which gave a high generalization performance.
3.7. Selection of critical product form features
When designing new products, designers often need to combine or re-organize
form features. Therefore, if the relative importance and ranking of the PFFs can be
analyzed, it can facilitate their decision making. Since the proposed PFFS
10
methodology is connected to the prediction model of CAR, not only the critical form
features can be selected but also their influence on producing specific affective
responses can be extracted. Whether the classification based methodology or
regression based PFFS is beneficial to the product development process depends on
the needs of the designers. In our previous research [42], an approach based on
support vector machine recursive feature elimination (SVM-RFE) is proposed to
streamline the selection of optimum PFFs. The same OVO multiclass fuzzy SVM as
proposed in [41] was used to construct a classification model for mobile phone design.
The results of our experiment show that SVM-RFE is not only very useful for
identifying critical form features with minimum generalization errors but also can be
used to select the smallest feature subset for building a prediction model with a given
discrimination capability.
4. Suggestions for future directions within the proposed framework
4.1. A consensus prediction model of affective responses
Most of the KES studies deal with the consumer’s data by averaging the
individual consumer evaluation data for the same product. The problem with this is
that the averaged data can only capture the most manifest trends in the consumer
satisfaction patterns. Since the CAR data are usually heterogeneous in nature,
averaged consumer data may fail to represent any individual opinion [43]. Instead of
constructing the prediction model of CAR using averaged SD evaluation data, we
propose a novel scheme called “consensus prediction model of affective responses”.
An overview of such a model is shown in Fig. 6. First, with the aid of a clustering
method, such as fuzzy clustering, applied to the consumer’s preference data
evaluating the representative product samples, all consumers can be grouped into
different clusters. The consumer group consisting of consumers S1 ,..., Sn can be
determined by the clustering algorithm. The relative importance w of each
consumer can be obtained using the distance measure d calculated from the
clustering algorithm. Secondly, a series of individual prediction models (with the SVR
algorithm say) are then constructed for consumers S1 ,..., Sn in the consumer group.
Typically, the product samples can be decomposed into PFFs x1 ,..., xm using
morphological analysis. Construction of these prediction models can be done by
taking x1 ,..., xm as input data while taking the individual CAR values y1 ,..., yn
toward every product samples as output values. Finally, the aggregation operator Ag
is used to integrate the individual CAR y1 ,..., yn of the SVR models SVR1 ,..., SVRn
to obtain an aggregated CAR value y Ag for each consumer group. Various
11
techniques can be used as an aggregation operator. A complete review of the existing
aggregation operators can be found in the paper by [44]. Consequently, the proposed
method is very suitable for modeling the CARs of consumers that exhibit
heterogeneous preference patterns while maintaining a high predictive performance
for product samples with unknown CARs.
< Insert Fig. 6 about here >
4.2. Affective response dimension selection using three-way data
Another interesting research topic within the proposed KES framework is how to
extend the ARDS methodology to deal with the differences among consumers.
Typically, the SD experiment results in a three-way data matrix to the order of
k  l  m , where k is the number of consumers, l the number of product samples
evaluated and m the number of adjectives on which the samples are evaluated. Note
that each consumer evaluates exactly the same product samples. In most of the KES
studies only the ( l  m ) two-way data matrix obtained by averaging over k
consumers’ ( l  m ) matrices is considered. Very few KES studies have been
conducted on the topic of analyzing the preservation of the multi-way format CAR
data. Yamamoto et al. [45] proposes a method for analyzing three-way SD data using
PA. However, their study emphasizes analyzing the differences among consumers
instead of selecting representative affective dimensions. In addition, existing methods
such as multiple factor analysis (MFA) [46] and parallel factor analysis (PARAFAC)
[47], developed in the field of sensory analysis, are worth further investigation.
4.3. Product form feature selection based on a consensus prediction model
According to the magazine article by Blum and Langley [48] feature selection
can be divided into three major categories: the filter, wrapper, and embedded methods,
based on the relationship between the selection scheme and the learning algorithm
used in the model’s construction. The embedded method directly incorporates the
feature selection process within the learning algorithm, thus often resulting in better
performance. The SVM-RFE algorithm used in the study of Shieh and Yang [42]
belongs to the embedded method. Since the prediction model is constructed according
to averaged CAR data across all subjects, the proposed PFFS method still fails to
consider the diversity and heterogeneity of the CAR data gathered from different
subjects. Therefore, an extension of our current research may establish the PFFS
method based on the same rationale for building the consensus prediction model
12
described in Sect. 4.1. Using the evaluated SD data, an individual prediction model
for each consumer in the same group should be constructed first. By analyzing the
criteria provided by the adopted feature selection method, the relative importance of
form features can be identified and a feature ranking for each consumer obtained.
However, such a consensus PFFS method should be capable of combining the feature
ranking obtained from the PFFS method, calculated from an individual prediction
model, instead of aggregating the output CAR value for an individual prediction
model. Therefore, different kinds of aggregation operators are needed to integrate the
ordinal ranking of the form features.
5. Concluding remarks
As a kind of consumer-oriented technology, KES can be used to systematically
study the subjective perceptions of consumers and establish an interrelationship with a
product’s design. Since the 1970’s, KES has attracted much attention from researchers
in the product design field, especially in Asia. However, the research framework
underlying KES has not improved much since its first appearance. In this study, a
general framework for KES is proposed to extend the work of Nagamachi’s [9] and
Schutte et al.’s [10]. It introduces CS, ARDS and PFFS. Important concepts related to
this proposed framework are described and a suitable methodology, which can be
used to implement the procedures of this framework, proposed. Preliminary results
from the authors’ recent studies are also summarized. Finally, we suggest several
future directions within the proposed framework, including a consensus prediction
model of affective responses, ARDS using three-way data, and PFFS based on a
consensus prediction model. This is the main research direction being taken by the
laboratory under the guidance of Meng-Dar Shieh. Nerveless, more case studies using
different kinds of products, such as consumer electronics, furniture, automobiles, etc.,
are still needed to verify the effectiveness of the proposed KES framework.
References
[1]
[2]
[3]
S.H. Han and S.W. Hong, A systematic approach for coupling user satisfaction
with product design, Ergonomics 46 (13/14) (2003) 1441-1461.
Y. Shimizu and T. Jindo, A fuzzy logic analysis method for evaluating human
sensitivities, International Journal of Industrial Ergonomics 15 (1995) 39-47.
M. Nagamachi, Kansei Engineering: a new ergonomic consumer-oriented
technology for product development, International Journal of Industrial
Ergonomics 15 (1995) 3-11.
13
[4]
S.W. Hsiao and H.C. Tsai, Applying a hybrid approach based on fuzzy neural
[8]
network and genetic algorithm to product form design, International Journal of
Industrial Ergonomics 35 (2005) 411-428.
S.W. Hsiao and E. Liu, A neurofuzzy-evolutionary approach for product
design, Integrated Computer-Aided Engineering 11 (2004) 323-338.
S.W. Hsiao and C.H. Chen, A semantic and shape grammar based approach for
product design, Design studies 18 (1997) 275-296.
J. Park and S.H. Han, A fuzzy rule-based approach to modeling affective user
satisfaction towards office chair design, International Journal of Industrial
Ergonomics 34 (2004) 31-47.
S.H. Han, K.J. Kim, M.H. Yun, and S.W. Hong, Identifying mobile phone
[9]
design features critical to user satisfaction, Human Factors and Ergonomics in
Manufacturing 14 (1) (2004) 15-29.
M. Nagamachi, Kansei Engineering, Kaibundo Publisher, Tokyo, 1993.
[5]
[6]
[7]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
S. Schutte, J. Eklund, J. Axelsson, and M. Nagamachi, Concepts, methods and
tools in Kansei engineering, Theoretical Issues in Ergonomics Sciences 5 (3)
(2004) 214-231.
S.H. Hsu, M.C. Chuang, and C.C. Chang, A semantic differential study of
designers' and users' product form perception, International Journal of
Industrial Ergonomics 25 (2000) 375-391.
J.M. Lamb and M.J. Kallal, A conceptual framework for apparel design,
Clothing & Textiles Research Journal 10 (2) (1992) 42-47.
K. Chen, Form language and style description, Design studies 18 (1997)
249-274.
M.C. Chuang and Y.C. Ma, Expressing the expected product images in product
design of micro-electronic products, International Journal of Industrial
Ergonomics 27 (2001) 233-245.
J.C. Jones, Design Methods, Van Nostrand Reinhold, New York, 1992.
J. Kwahk and S.H. Han, A methodology for evaluating the usability of
audiovisual consumer electronic products, Applied Ergonomics 33 (2002)
419-431.
K. Chen, A study of computer-supported formal design, Design studies 19
(1998) 331-359.
D.R. Wallace and M.J. Jakiela, Automated product concept design: unifying
aesthetics and engineering, IEEE Computer Graphics and Applications 13 (4)
(1993) 66-75.
H.H. Lai, Y.C. Lin, C.H. Yeh, and C.H. Wei, User-oriented design for the
optimal combination on product design, International Journal of Production
14
Economics 100 (2006) 253-267.
[20]
[21]
[22]
[23]
[24]
D. MacKay, Chemometrics, econometrics, psychometrics—How best to
handle hedonics?, Food Quality and Preference 17 (2006) 529-535.
S.W. Hsiao and H.C. Huang, A neural network based approach for product
form design, Design studies 23 (2002) 67-84.
V.N. Vapnik, The nature of statistical learning theory, Springer, New York,
1995.
C. Burges, A tutorial on support vector machines for pattern recognition, Data
Mining and Knowledge Discovery 2 (2) (1998) 955-974.
W. Smith, Product differentiation and market segmentation as an alternative
marketing strategy, Journal of Marketing 21 (1) (1956) 3-8.
[25]
T. Bock and M. Uncles, A taxonomy of differences between consumers for
market segmentation, International Journal of Research in Marketing 19 (2002)
215-224.
[26]
H.W. Shin and S.Y. Sohn, Segmentation of stock trading customers according
to potential value, Expert Systems with Applications 27 (2004) 27-33.
C.E. Osgood, C.J. Suci, and P.H. Tannenbaum, The Measurement of Meaning,
University of Illinois Press, Champaign, IL, 1957.
P. Coxhead and J.M. Bynner, Factor analysis of semantic differential data,
Quality and Quantity 15 (1981) 553-567.
K.A. Hsiao and L.L. Chen, Fundamental dimensions of affective responses to
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
product shapes, International Journal of Industrial Ergonomics 36 (2006)
553-564.
H. You, T. Ryu, K. Oh, M.H. Yun, and K.J. Kim, Development of customer
satisfaction models for automotive interior materials, International Journal of
Industrial Ergonomics 36 (2006) 323-330.
E. Alcantara, M.A. Artacho, J.C. Gonzalez, and A.C. Garcia, Application of
product semantics to footwear design. Part I—Identification of footwear
semantic space applying differential semantics, International Journal of
Industrial Ergonomics 35 (2005) 713-725.
M.C. Chuang, C.C. Chang, and S.H. Hsu, Perceptual factors underlying user
preferences toward product form of mobile phones International Journal of
Industrial Ergonomics 27 (2001) 247-258.
K. Sahmer and E.M. Qannari, Procedures for the selection of a subset of
attributes in sensory profiling, Food Quality and Preference 19 (2008)
141-145.
E. Vigneau and E.M. Qannari, Clustering of variables around latent
components, Communications in Statistics-Simulation and Computation 32 (4)
15
(2003) 1131-1150.
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
W.J. Krzanowski, Selection of variables to preserve multivariate data structure,
using principal components, Applied Statistics 36 (1) (1987) 22-33.
T. Wakaki, H. Itakura, and M. Tamura, Rough set-aided feature selection for
automatic web-page classification, in: Proceedings of the IEEE/WIC/ACM
International Conference on Web Intelligence, pp. 70-76, 2004.
S.K. Pal, J. Basak, and R.K. De, Feature selection: a neuro-fuzzy approach, in:
IEEE International Conference on Neural Networks, Washington, DC, pp.
1197-1202, 1996.
Y. Liu and Y.F. Zheng, FS_SFS: A novel feature selection method for support
vector machines, Pattern Recognition 39 (2006) 1333-1345.
L. Hermes and J.M. Buhmann, Feature selection for support vector machines,
in: 15th International Conference on Pattern Recognition, pp. 2712, 2000.
C.C. Yang and M.D. Shieh, Selecting representative affective dimensions for
product design using factor analysis and Procrustes analysis, Preparing
manuscript (2008).
M.D. Shieh and C.C. Yang, Classification model for product form design
using fuzzy support vector machines, Computers & Industrial Engineering
(2008) doi:10.1016/j.cie.2007.12.007.
M.D. Shieh and C.C. Yang, Multiclass SVM-RFE for product form feature
selection,
Expert
Systems
with
Applications
(2007)
doi:10.1016/j.eswa.2007.07.043.
M.D. Lee and K.J. Pope, Avoiding the dangers of averaging across subjects
when using multidimensional scaling, Journal of Mathematical Psychology 47
(2003) 32-46.
Z.S. Xu and Q.L. Da, An overview of operators for aggregating information,
International Journal of Intelligent Systems 18 (2002) 953-969.
K. Yamamoto, T. Yoshikawa, and T. Furuhashi, Division method of subjects
by individuality for stratified analysis of SD evaluation data, in: IEEE
International Symposium on Micro-NanoMechatronics and Human Science,
pp. 29-34, 2005.
S. Le and S. Ledauphin, You like tomato, I like tomato: segmentation of
consumers with missing values, Food Quality and Preference 17 (2006)
228-233.
N.D. Sidiropoulos, R. Bro, and G.B. Giannakis, Parallel factor analysis in
sensory array processing, IEEE Transaction on Signal Processing 48 (8) (2000)
2377-2388.
A.L. Blum and P. Langley, Selection of relevant features and examples in
16
machine learning, Artificial Intelligence 97 (1997) 245-271.
17
Fig. 1. Conceptual model of KES (modified from [1]).
18
Fig. 2. Morphological analysis based PFF representation [6].
19
Fig. 3. Object-oriented PFF representation [16].
20
Fig. 4. Level-of-detail PFF representation [18].
21
Fig. 5. A seven-step general framework for KES.
Step1: Selection of representative products
Step2: Selection of representative adjectives
Step3: Segmentation of consumers
Step4: Determination of product form
representation
Step5: Experimental and questionaire design
Step 6: Construction of prediction model for
affective responses
Step7: Selection critical product form features
22
Fig. 6. Overview of the consensus prediction model of affective responses.
SVR1
SVRn
SVR-consumer S n
yn 1
yn
Product samples
morphological
analysis
w2
d 2,n 1
importance of
consumers
Aggregated
w
response
y Ag
representative
products
23
d1,n
Sn
wn
d1,n 1
d n , n 1
wn 1
preference
evaluation
S n 1
oup
SVR-consumer Sn 1
d 2,n
w1
other gr
SVRn 1
S2
p
ou
Aggregation
operator
Ag
...
Input
form
feautures
x1 ,..., xm
gr
SVR-consumer S 2
Consumer
group
S1
y2
er
SVR2
Fuzzy clustering
h
ot
SVR-consumer S1
individual
response
y1
Table 1. Aesthetic adjectives adapted from [13].
1 Harmonious-contrasting
2 Homogeneous-heterogeneous
3 Geometric-biomorphic
4 Pure-impure
5 Simple-complex
6 Balanced-unstable
24
7 Monolithic-fragmentary
8 Static-dynamic
9 Uniform-multiform
10 Functional-decorative
11 Subtle-bold
12 Single-multiple
Table 2. Expressive adjectives adapted from [11].
1 Traditional-modern
2 Hard-soft
3 Old-new
4 Heavy-handy
5 Obedient-rebellious
6 Nostalgic-futuristic
7 Coarse-delicate
8 Masculine-feminine
9 Rational-emotional
10 Hand made-hi tech
11 Childish-mature
12 Unoriginal-creative
13 Simple-complicated
14 Conservative-avant grade
15 Standard-outstanding
16 Common-particular
17 Plain-gaudy
18 Decorative-practical
25
19 Inert-active
20 Personal-professional
21 Obtuse-brilliant
22 Discordant-harmonious
Table 3. An example of PFFs for mobile phone design.
Form features
Type
Attributes
Length
Continuous
None
Continuous
None
Continuous
None
Continuous
None
(X1)
Width
(X2)
Thickness
Body
(X3)
Volume
(X4)
Type
Discrete
(X5)
Type
Block body
Flip body
Slide body
(X5-1)
(X5-2)
(X5-3)
Full-
Partial-
Regular-
separated
separated
separated
(X6-1)
(X6-2)
(X6-3)
Round
Square
Bar
(X7-1)
(X7-2)
(X7-3)
Circular
Regular
Asymmetric
(X8-1)
(X8-2)
(X8-3)
Square
Vertical
Horizontal
(X9-1)
(X9-2)
(X9-3)
Discrete
Function button
(X6)
Style
Discrete
(X7)
Shape
Discrete
Number button
(X8)
Arrangement
Discrete
(X9)
26
Detail treatment
Discrete
(X10)
Regular seam
Seamless
Vertical seam
Horizontal
(X10-1)
(X10-2)
(X10-3)
seam
(X10-4)
Position
Discrete
Panel
(X11)
Shape
Middle
Upper
Lower
Full
(X11-1)
(X11-2)
(X11-3)
(X11-4)
Square
Fillet
Shield
Round
(X12-1)
(X12-2)
(X12-3)
(X12-4)
Discrete
(X12)
27
Related documents
Download