Virtual Concept 2003   Biarritz - France   November, 5-7  1  SUBJECTIVE EVALUATION OF FORMS IN AN IMMERSIVE ENVIRONMENT  Jean-François Petiot, Damien Chablat Ecole Centrale de Nantes IRCCyN – UMR CNRS 6597 – Equipe MCM 1, rue de la Noë BP 92101 44321 NANTES Cedex 3 France Phone : 33 2 40 37 69 59 – Fax : 33 2 40 37 69 30 E-mail : {Jean-Francois.Petiot, Damien.Chablat}@irccyn.ec-nantes.fr  Abstract :   User’s perception of product, by essence subjective, is a major topic in marketing and industrial design. Many methods, based on users’ tests, are used so as to characterise this   perception.   Methods   like   multidimensional   scaling, semantic differential method, and preference mapping are well known in sensorial analysis or in the food industry. These methods are used in order to built a perceptual space in order to position a new product, to specify requirements by the study of user’s preferences, to evaluate some product attributes, related in particular to style (aesthetic). These early stages of the design are essential for a good orientation of the project. In parallel, virtual reality tools and interfaces are more and more efficient for suggesting to the user complex feelings, and creating in this way various levels of perceptions. In this article,   we   present   on   an   example   an   adaptation   of multidimensional   scaling   and   preference   mapping   for   the subjective   assessment   of   virtual   products.   These   products, which geometrical form is variable, are defined with a CAD model and are proposed to the user with a spacemouse and stereoscopic glasses. Advantages, uses and limitations of such evaluation is next discussed.  Key words   : subjective evaluation, multidimensional scaling, virtual reality, preference mapping.  1- Introduction  In today’s highly competitive market, developing new products that meet consumers’ tastes is a crucial issue in product design. To improve attractiveness, a well-designed product should not only satisfy requirements, but should also satisfy consumers’ psychological   needs,   by   essence   subjective.   For   many products, the aesthetic properties are as important as technical functions [1]. In order to predict the success of a product, to control and to optimise its performances, it’s time now to take into account esteem and aesthetics functions in the beginning of the design. Although industrial and product designers are keenly aware of the importance of design aesthetics [2], there is an obvious lack of systematic, scientific, and engineering methods to help them make   aesthetic   design   decisions   and   conduct   aesthetic evaluation. For example, the perception of the shape of a product is often nothing but a style of design, depending much more on the designer’s taste than on real customers’ trends, as some studies clearly showed [3]. In   this   context,   Virtual   Reality   (VR)   seems   to   offer promising   functionalities   for   the   assessment   of   virtual products [4]. The Virtual reality interfaces available on the market are now mature enough for suggesting to the user relevant feelings and sensations [5]. The main problem is now to learn how to use it and to define relevant methods for their integration into the design process. How do we use virtual products to study the aesthetics of products and to help   make   them   more   aesthetic   ?   How   increase   the attractiveness of the shape of a product via virtual reality tools ? In this article, we present the use of VR tools and user-tests for the study of aesthetics. It is assumed that we possess an innate   ability   to   perceive   a   wide   range   of   qualities   in products that shape our response to them [6]. But it’s often more difficult to detect bad proportions in the design of products, or to explain rules which lead to a feeling of harmony and beauty. A study based on the golden section or “divine proportion”, described in [7], shows that it is not always clear whether the artists consciously used the golden section   or   whether   they   intuitively   approximated   the associated ratio. Aesthetics is clearly a multidimensional characteristic, not independent of ergonomics and usability considerations. The concept of ergo-aesthetics, proposed in [8], refer to the integrated design approach that is aimed at meeting both ergonomic   and   aesthetic   design   objectives.   Two   main approaches can be used for the study of aesthetics [8]: -   The top-down approach attempts to understand aesthetic responses as a whole, -   The bottom-up approach attempts to identify the basic pictorial features and compositional patterns that please or displease the senses.
Virtual Concept 2002   Biarritz - France   October, 9-10  2 In this article, we propose an illustration of both approaches for the evaluation of forms in   an immersive   environment.   In section   2,   we   present   two   classical   methods   in   sensorial analysis which can be used in product design for grasping the subjective response of subjects. In section 3, we propose an adaptation of these methods to the experimental study of the “appeal” of forms. The perception of the aesthetic of forms by a   subject   is   carried   out   in   a   VR   environment,   and   with particular products: table glasses. The main results of this experiment are discussed in section 4, and perspectives for the use of VR for the design of forms are discussed.  2- Background  To study users’ perceptions, researchers in marketing and sensorial   analysis   propose   various   methods   [9][10][11]. Perceptual maps are commonly used to take perceptions into account and to control the product positioning. The basic idea is to build a multi-attribute perceptual space in which each product is represented by a point. We propose to describe  multidimensional   scaling   (MDS),   a   classical   method   for building   the   perceptual   space,   and   preference   mapping  techniques, which are relevant to grasp preference assessments. An adaptation of both methods will be used in section 3.  2.1 – MDS  Multidimensional   scaling   uses   dissimilarity   assessments   to create a geometrical representation of a family of “stimuli”. This method, developed initially for psychometric analysis [12], is a process whereby a distance matrix among a set of stimuli is translated into a representation of these stimuli inside of a perceptual space. Taking all the possible pairs of stimuli (here pairs of products) into account, the subject evaluates their degree of similarity on a quantitative scale and fills in this way a dissimilarity matrix. Technically, what MDS does is to find a set of points   (X i ) i=1,...,N   in a K-dimensional space such that the distances among them correspond as closely as possible to the dissimilarity (or a function of it) in the input matrix, according to a given optimisation criterion. This criterion can be for example a function called   stress,   (equation 1) which represents the “badness of fit”, and the distance can be the Euclidean distance (equation 2), but several different expressions can be used [13].   D   ij   is the perceptual dissimilarity between stimuli i and j,   d   ij   the Euclidean distance,   x ik   the coordinate of stimuli   i  on dimension   k ,   K   the number of dimensions of the perceptual space, selected by the experimenter. After computation, each stimuli is represented in the Euclidean space by a point   X i (x i1 , ...,x iK   ) . 2 / 1 2 2 ) (                − = ∑∑ ∑∑  i   j ij i   j ij ij d D d stress  (1)   ∑ = − =  K k jk ik ij   x x d  1 2 ) ( (2) The main advantage of this method is that the tests are based on instinctive dissimilarity assessments, which do not impose any criteria or predefined semantic scale. This method provides a space for a visualisation of the relevant dimensions for stimuli’s perception. It is used in marketing and sensorial analysis,   the   stimuli   can   be   sounds,   fragrance,   products, concepts,...It is well suited to study the relationship between products. We have used in the next section metric MDS, based on the minimisation of the stress, and using a sparse matrix of dissimilarity.  2.1 – Preference mapping  Preference   mapping   constitutes   a   group   of   statistical techniques which are also common in sensory analysis and marketing   research.   Internal   (MDPREF)   and   external (PREFMAP)   preference   mapping   both   generate   graphic interpretations   of   individual   preferences   [14][15].   To construct   such   preference   mappings,   multiple   regression techniques are used. The principle is to find a correlation between product’s preference and product’s position in the perceptual space. Technically speaking, one must perform a multiple linear regression of the preference on the perceptual coordinates.   These   techniques   need   some   strong assumptions. The preference’s model is supposed to be linear most of the time, it’s not possible to consider threshold phenomena for the preference, and the method needs an absolute   evaluation   of   the   preference   as   input   (hedonic scale). The use of a hedonic scale to assess the preference in an   absolute   manner   encounters   some   difficulties   and limitations which can be avoided by the use of pairwise comparison   methods   [16].   In   the   PREFMAP   model, consumers   are   represented   by   their   ideal   point   in   the perceptual   space,   which   represents   the   most   preferred combination   of   attributes.   Other   models   with   infinite preferred attribute levels are called ‘vector models’. We use in the next section preference mapping based on the vector model.  3- Experimental study  3.1 – Description of the VR tools  The main objective of the experiment is to show that the study of the aesthetic properties of forms can be conducted on virtual products, and that an immersive environment is particularly helpful for the assessment and the improvement of aesthetics. To illustrate our approach, we propose to study the aesthetics of table glasses, which are very interesting products from an aesthetics   and   esteem   point   of   view.   A   study   on   such products (wine–glass) is proposed in [17], where the authors present a method for form generation, and in [18], where a methodology for a solid assessment of product semantics is proposed. A digital model of glasses with various forms was designed with the CAD and VR software   Catia V5R11 . A texture with the material “glass” were applied to the models of glass. A Spacemouse and stereoscopic eyeglasses were used by the subject for enhancing his/her perception of the environment, and for facilitating the various assessments.  3.2 – Description of the assessment tests  Several forms of table glasses were generated with the CAD software. All glasses are intended to the same usage, for example wine glasses. They all have the same general form, balloon glass, and are made off three parts: base, foot and
Virtual Concept 2002   Biarritz - France   October, 9-10  3 container.   The   mathematical   model   of   the   curve   for   the generation of forms is given figure 1. The base, the foot, and the container are modelled with a spline-curve with 15 control nodes. Continuity conditions are added at points P5 and P9. The inner shape of the container is parallel to the outer shape, with a gap of 1 mm. The 3D solid is generated by rotation of the curve. P 1 P   2 P   3 P 4 P 5 P   6 P   7 P 8 P   9   P   10 P   11 P   12 P   13 y x  Fig. 1   : Model of the outer curve of the glass.  #1   #2   #3   #4  #5   #6   #7   #8  #9   #10   #11   #12  #13   #14   #15   #16  #17   #18  Fig. 2   : Pictures of the 18 glasses. All   the   glasses   used   for   the   experimental   study   are   an “instanciation” of this model. 18 glasses were generated (figure 2), which constitute the stimuli for the assessment tests. VR tools were next used for the evaluation of the aesthetics qualities of the glasses The assessment performed by the subject   was   made   of   several   stages,   presented   in   the following sections.  3.2.1 –Stage 1: Building of the perceptual space  The aim of this stage is to find how many dimensions can describe the aesthetic, or the “appeal” of the glasses, and to find   the   location   of   the   glasses   according   to   these dimensions.   For   that,   the   subject   had   to   quantify   the perceived differences between pairs of products from an aesthetic point of view. This stage belongs to the “top-down” approach. The subject was asked to say to what degree the products were different. He/she chose between “identical”, ”a little different”, ”different” and “very different”. The answer was coded on an integer scale from 0 (identical) to 3 (very different). In order to reduce the assessment task, the subject was not forced to fill all the dissimilarity matrix. He only had to chose the more obvious comparisons. He had the complete choice   of   the   comparisons   to   fill   in   the   N.(N- 1 )/ 2 = 153 potential comparisons of the superior half of the dissimilarity matrix. Nevertheless, so as to have computable data, and to reveal the multidimensionality of aesthetics, each product should be involved in at least three comparisons. The pairwise comparison were performed by the subject with the VR software and easy “click and drag” operations. The (sparse) dissimilarity matrix provided by the subject is given table 1. 1   2   3   4   5   6   7   8   9   10   11   12   13   14   15   16   17   18 1   0   2   *   1   *   *   0   *   *   *   1   *   *   *   2   *   *   * 2   2   0   1   3   2   1   2   0   1   3   2   1   1   0   1   3   2   1 3   *   1   0   2   2   *   *   *   0   2   2   *   *   2   0   *   3   * 4   1   3   2   0   2   2   1   3   2   0   1   3   2   3   2   1   2   2 5   *   2   2   2   0   *   1   *   *   *   0   *   *   *   2   *   *   * 6   *   1   *   2   *   0   *   *   *   *   *   *   *   *   2   *   *   * 7   0   2   *   1   1   *   0   2   *   *   3   2   0   2   0   *   *   * 8   *   0   *   3   *   *   2   0   *   *   3   *   *   *   1   *   *   * 9   *   1   0   2   *   *   *   *   0   *   3   *   *   *   0   *   *   * 10   *   3   2   0   *   *   *   *   *   0   1   *   *   *   2   *   *   * 11   1   2   2   1   0   *   3   3   3   1   0   1   2   3   3   1   0   3 12   *   1   *   3   *   *   2   *   *   *   1   0   *   *   2   *   *   * 13   *   1   *   2   *   *   0   *   *   *   2   *   0   *   2   *   *   * 14   *   0   2   3   *   *   2   *   *   *   3   *   *   0   2   *   *   * 15   2   1   0   2   2   2   0   1   0   2   3   2   2   2   0   2   1   1 16   *   3   *   1   *   *   *   *   *   *   1   *   *   *   2   0   *   * 17   *   2   3   2   *   *   *   *   *   *   0   *   *   *   1   *   0   * 18   *   1   *   2   *   *   *   *   *   *   3   *   *   *   1   *   *   0  Tab. 1   : Sparse dissimilarity matrix. Next, we used MDS for determining how many dimensions can   represent   the   dissimilarities,   and   for   building   a perceptual space which expresses these differences. With this sparse matrix as input, an own implementation of metric   MDS   has   been   used   to   calculate   the   perceptual coordinates   of   the   glasses   [19].   A   2-dimensional configuration, with a stress value equal to 0.12 (considered as a correct “badness of fit”) has been retained (figure 3).
Virtual Concept 2002   Biarritz - France   October, 9-10  4  -1   -0.5   0   0.5   1   1.5   2   2.5 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 #1 #2 #3 #4 #5 #6 #7 #8 #9 #10 #11 #12 #13 #14 #15 #16 #17 #18 axe 1 axe 2 2 D-perceptual space  Fig. 3   : 2D Perceptual space of the glasses These results reveal the two main dimensions (axe1, axe2) which are relevant for describing the differences between the products from an “appeal” point of view.  3.2.2 –Stage 2: assessment of the appeal  The aim of this stage is to get an absolute assessment of the appeal, on a continuous scale. The task is the same as previous stage, but the subject is now forced to provide a mono- dimensional answer. In order to reveal the multidimensionality of aesthetic during stage 1, it’s important that the subject do not carry out stage 1 from the results of stage 2. In other words, stage 1 must be performed before stage 2. The subject was asked to rank the products in the order of increasing appeal, or into piles of similar appeal. The subject associated to each product (or each pile) an absolute value of the appeal, on a continuous scale from 0 to 10. This scale is similar to the “hedonic” evaluation used in the food industry for preference assessments. The assessment of the glasses, provided by the subject, is given table 2. Appeal P 0   1   2   3   4   5   6   7   8   9   10 Glass #   2   8 14 12 18 9 15 6 3 17   13 7 1   11 5 16   4 10  Tab. 2   : Appeal scores of the glasses.  3.2.3 –Stage 3: a model of appeal  The main objective is to find how sensitive is the subject in detecting small variations in aesthetics, and to find his abilities to perceive and judge values, changes, and variations in design parameters.   This   stage   belongs   now   to   the   “bottom-up” approach [20]. For that, three shape regulating rules were proposed to the subject. These rules correspond to a modification of the three determining design parameters of a glass, proposed in [21]. They are expressed as following (figure 4): -   R1: increase the total height of the container, d1, -   R2: increase the height of the foot, d2, -   R3: increase the diameter of the container, d3. R1   R2   R3 d1 d2 d3  Fig. 4   : Description of the shape regulating rules For each glass Gj, the subject has to indicate the effect of the proposed shape regulating rule on his/her assessment of the appeal: -   If rule “Ri” increases the appeal,   ∆ Pi(Gj) = 1 -   If rule “Ri” does not change the appeal,   ∆ Pi(Gj) = 0 -   If rule “Ri” decreases the appeal,   ∆ Pi(Gj) = -1 This is equivalent to an assessment of the “derivative” of the appeal according to the design parameters. The virtual reality model of the glass is particularly suitable for the required assessments. Indeed, the subject can even make the modification on the virtual model to visualise and compare the differences, if the assessment of the appeal change is not obvious. The values of the dimensions d1, d2, d3 for each glass and the assessments proposed by the subject are given in the following table 3. For each shape regulating rule, these values   ∆ Pi(Gj)   are   proportional   to   the   partial   derivative  dj P  ∂ ∂   , with a positive multiplicative factor kj. Glass # d1 cm d2 cm d3 cm R1 1 1   d P k   ∂ ∂  R2 2 2   d P k   ∂ ∂  R3 3 3   d P k   ∂ ∂  G1   8   7   8   -1   0   -1 G2   8   7   9,5   -1   1   -1 G3   8   7   9,5   -1   0   -1 G4   8   7   6   -1   -1   0 G5   8   7   7   -1   -1   -1 G6   8   7   9   -1   0   -1 G7   8   5   8   -1   1   -1 G8   8   5   9,5   -1   1   -1 G9   8   5   9,5   -1   1   -1 G10   8   5   6   -1   1   0 G11   8   5   7   -1   1   -1 G12   8   5   9   -1   1   -1 G13   8   3   8   -1   0   -1 G14   8   3   9,5   -1   1   -1 G15   8   3   9,5   -1   1   -1 G16   8   3   6   -1   1   -1 G17   8   3   7   -1   1   -1 G18   8   3   9   -1   1   -1  Tab. 3   : Dimensions of the glasses and assessment of the appeal changes according to the shape regulating rules.
Virtual Concept 2002   Biarritz - France   October, 9-10  5  3.2.3 –Building of an model of appeal  A quadratic model of the appeal P is proposed. The predicted value, denoted   P   , is given by: 10 3 . 2 . 9 3 . 1 . 8 2 . 1 . 7 3 . 6 2 . 5 1 4 3 . 3 2 . 2 1 . 1   2 2 2  a d d a d d a d d a d a d a d a d a d a d a P  + + + + + + + + + =  (3) The predicted partial derivatives of P are given by: 3 ). 2 . 9 1 . 8 6 . 2 ( 3 3 2 ). 3 . 9 1 . 7 5 . 2 ( 2 2 1 ). 3 . 8 2 . 7 4 . 2 ( 1 1  d d a d a a a d P d d a d a a a d P d d a d a a a d P  + + + = ∂ ∂ + + + = ∂ ∂ + + + = ∂ ∂  (4) The set of coefficients {a1, a2,..., a10} is determined with an optimisation procedure, similar to the least square regression. It’s a generalisation of the linear regression of the appeal on the   design   parameters,   taking   into   account   more   data concerning the partial derivative. With   all observations   ) , (   dj P P   i i   ∂ ∂   ,   the vector of   variables  X=   (a1, a2,..., a10, k1, k2, k3) is determined by minimising the following function   F   : ) , , , ,..., , ( . . . ) ( ) ( ) ( min 3 2 1 10 2 1 2 1 3 1 2 1  k k k a a a X t r w d P k d P P P X F   N i   j i j j   j i N i i i  = ∂ ∂ − ∂ ∂ + − =   ∑ ∑ ∑   =   = =   (5)  i P   is the appeal,   dj P i  ∂ ∂   the partial derivative, provided by the subject, for glass G   i ,  i P   is the appeal,   dj P   i  ∂ ∂   the partial derivative, provided by the model, for glass G   i , The result of the optimisation procedure is given table 2. a1   a2   a3   a4   a5   a6   a7   a8   a9   a10 3.6   0.12   13.6   -1.52   0.02   -0.13   -0.01   -1.71   0.01   82.36 k1   k2   k3 -0.01   14.15   0.01  Tab. 4   : Coefficients of the model of appeal. The response surface, corresponding to this model, is plotted in the next section.  4- Results and discussions  4.1 – Preference mapping 1  The first analysis which can be done is to try to explain the appeal of the glasses by the perceptual dimensions given by MDS. Like for the PREFMAP model, this is done by a multiple regression   using   the   perceptual   axes   (axe1,   axe2)   as independent variables and the appeal P as the dependent variable (equation 4).  c axe b axe a P   + + =   2 . 1 .   (6) The results of the regression is given table 5. a   b   c   R 2   F   Significant 1   (p=0.01) -2   -1,8   7   0,91   80   yes  Tab. 5   : Coefficients of the model of preference mapping 1 The   regression   is   significant   and   the   determination coefficient R 2   is close to 1. So the linear model is well adapted for describing the data provided by the subject. This good correlation shows a good coherency of the subject: the data provided in stage 1 are quite compatible with the data provided in stage 2. Furthermore, the VR interfaces proposed to the subject do not introduce incoherence or “noise” in the assessment of the shape. In other words, the good coherence of the data is a sign for the perfect adaptation of the interface to the task proposed to the subject. So as to give a graphical interpretation of the regression, the vector model of the appeal is plotted in the perceptual space. The origin of the vector is located arbitrarily in the origin of the frame, the values of the regression coefficients (a, b) give the orientation of the arrow, the arrowhead points in the direction of increasing appeal (figure 5). It can be shown that this vector is parallel to the steepest slope line of the plane (equation (6)). The perpendiculars to the vector are the iso- appeal curves.  -1 0 1 2 -1   -0,5   0   0,5   1   1,5   2  Appeal  2D-perceptual space  16 11 6 15 14 1 2 3 4 5 13 7 8 9 10 12 17 18  Fig. 5   : vector model of appeal in the perceptual space 1   according to a Fisher-Snedecor table
Virtual Concept 2002   Biarritz - France   October, 9-10  6 The interest of this graph   lies in its ability to   study the multidimensionality of aesthetics. What differentiate glass #13 and #17, which have about the same appeal ? May a particular attribute,   or   a   particular   design   parameter,   explain   the difference ? Further studies are needed for answering these questions.  4.2 – Preference mapping 2  The second analysis which can be done is to try to explain the appeal of the glasses by certain design parameters of the forms. The response surface of the appeal   P   , obtained in section 3, can be plotted, according to the design variables (d2, d3) (equation 7) 2:  d3 d2 d3 d2 d3 d2  . . 9 . 6 . 5 ). 8 . 8 3 ( ). 7 . 8 2 ( 4 . 64 10 1 . 8 2 2   a a a a a a a a a a P  + + + + + + + + + =  (7) The colormap of this surface is presented figure 6.  Fig. 6   Colormap of the appeal in the design parameter space (d2, d3). The structure of the appeal seems to be very simple and uni- dimensional.   Increasing   d2   and   decreasing   d3   lead   to   an increase   of   the   appeal.   The   perpendicular   to   the   steepest descent line, given by the relation d3=d2/2+c, is about an iso- appeal line (figure 7). 2   for all glasses, the dimension d1=8cm  5 6 7 8 9 2   3   4   5   6   7   8  d2 d3  16 11 6 15 14 1 2 4 5 13   7 8 9 10 12 17 18 3  Appeal Iso appeal line  Fig. 7   : Iso-appeal line in the design parameter space The relation d3=d2/2+c is important for the study of the aesthetic of a glass. It’s a characteristic of the subject, who has provided data which satisfy this relation. The strong point of our study is that the methodology presented allows the discovery of such relation. The subject is in most of the cases unconscious of this relation and is not able to formulate it explicitly. Both preference mappings are interesting for improving the design of forms, or seeking “good” products because they use two complementary approaches.  5- Conclusions  We have presented in this paper an experimental study of the aesthetic of table glasses. This study of the “appeal” of the shapes has been carried out on virtual models, generated by a CAD/VR   software,   and   with   classical   VR   interface:   a spacemouse and stereoscopic eyeglasses. With this tools, the study was based on the assessment by a subject of various characteristics of the glass. Two ways of investigations have been used: (1) a top-down approach, using multidimensional scaling and linear regression of the “appeal” of the shapes ; (2) a bottom-up approach, with the building of a model of the appeal, based on the assessment of “shape regulating rules “. A new method for establishing the model of appeal, based on an optimisation process, has been proposed. Firstly, the good coherence of the data proposed by the subject shows the perfect adaptation of the interfaces to the proposed task. Furthermore, the subject does not encounter difficulties in the assessment tasks. Secondly,   the   model   of   appeal,   based   on   two   design parameters of the shape of the glass, reveals how these parameters are linked for the subject’s assessments. In perspective, we will conduct a similar study using more design   parameters,   and   based   on   principal   component analysis. The next step will be to use a model of subjective evaluation   for   forms   generation   and   product   synthesis, approach used for example in Kansei engineering type 3
Virtual Concept 2002   Biarritz - France   October, 9-10  7 [22]. In future works, haptic devices will be introduced for the assessment of weight, stability and more generally ergonomic attributes.  5- Acknowledgements  We acknowledge the help of Hélène Compain for performing the tests and for various assessments of aesthetics.  6- Bibliography  [1] Wagner R. “Aesthetics Design Improvement”. International Conference on Engineering Design ICED 01. Glasgow, August 21-23 2001. pp389-396. [2] Quarante D., (2001), Eléments de design industriel (3ème édition). Polytechnica. [3]   Hsu   S.H.,   Chuang   M.C.,   Chang   C.C. « A   semantic differential   study   of   designers’   and   users’   product   form perception ». International Journal of Industrial Ergonomics. 25 (2000), 375-391. [4] D. Paillot, F. Merienne, J-P. Frachet, M. Neveu, S. Thivent, L. Fine. Revue de projet immersive pour le style automobile. Proceedings of Virtual Concept 2002, Biarritz, France, pp 92- 97, 2002. [5] X. Fischer, N. Troussier. La réalité virtuelle pour une conception centrée sur l’utilisateur. Proceedings of Virtual Concept 2002, Biarritz, France, pp 80-89, 2002. [6] Macdonald A.S. “Aesthetic intelligence: optimizing user- centered design” Journal of Engineering Design. Vol. 12., N°1 37-45. [7] Lipovetsky S., Lootsma F.A. Generalized golden sections, repeated bisections and aesthetic pleasure. European Journal of Operational Research 121, 213-216, 2000. [8] Liu   Yili.,   Engineering   Aesthetics   and   Ergo-Aesthetics: Theoretical and Methodological Foundations. Proceedings of the   5   th   Annual   International   Conference   on   Industrial Engineering Theory, Applications and Practice, 2000. [9] Kaul A., Rao R.V. Research for product positioning and design decision: an integrative review. International Journal of Research in Marketing 12, 293-320, 1995. [10]   Evaluation   Sensorielle,   manuel   méthodologique,   2 ième édition. Techniques et Documentation Lavoisier, collection Sciences et Techniques agro-alimentaires, 1998. [11]   Y.   Evrard,   B.   Pras,   E.   Roux.   MARKET :   Etudes   et Recherches en Marketing. DUNOD 2000. [12]   Shepard   R.   N.,   Romney   K.,   Nerlove,   S.   B.   (Eds.) “Multidimensional scaling: Theory   and   applications   in   the behavioral sciences”, Volume I: Theory.   New York: Seminar Press, 1972. [13]   Davison   M.L.,   Multidimensional   Scaling .   Krieger Publishing Company, Malabar, Florida, 1992. [14] Caroll J.D., Individual differences and multidimensional scaling. In   Multidimensional scaling: theory and application in the behavioral sciences , edited by Shepard, Romney, Nerlove, Vol 1, 105-155, 1972. [15] ECRIN. Evaluation Subjective : Méthodes, Applications et Enjeux. Les cahiers du club CRIN. 1997. [16] David H.A.,   The method of paired comparisons ,   (New York. Oxford University Press), 1988. [17]   Matsuoka   Y.,   Method   for   constructing   an   inverse reasoning system for form generation from product image objectives. Proceedings of ICED99, Vol.1, Munich, 137-142, 1999. [18] Petiot J-F., Yannou B. How to comprehend and assess product semantics: a proposal for an integrated methodology. Proceedings of the ICED 03, Stockholm, August 2003. [19]   Petiot J-F., Poirson E., Le Carrou J-L., Gilbert J. Differentiation of trumpets’ sounds : experimental study with an adaptable depth mouthpiece. Proceedings of Stockholm Musical Acoustic Conference SMAC03. Stockholm 2003. [20]   Hsiao   S.W.,   Wang   H.P.,   Applying   the   semantic transformation   method   to   product   form   design.   Design Studies 19, 309-330, 1998. [21] Trocherie R., (2002), Conception de produits : analyse de la sémantique produit et modélisation des préférences clients. Mémoire de DEA Génie mécanique, Ecole Centrale de Nantes. [22]   Nagamachi M., Kansei engineering: a new ergonomic consumer-oriented   technology   for   product   development. International Journal of Industrial Ergonomics, 15, 3-11, 1995.