Design facial appearance for roles in video games Expert Systems with Applications 36 (2009) 4929–4934 Contents lists available at ScienceDirect Expert Systems with Applications j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / e s w a Design facial appearance for roles in video games Shang Hwa Hsu *, Ching-Han Kao, Muh-Cherng Wu Department of Industrial Engineering and Management, National Chiao Tung University, 1001 Ta Hsueh Road, HsinChu 30010, Taiwan, ROC a r t i c l e i n f o Keywords: Video game Role Fuzzy AHP BP neural network 0957-4174/$ - see front matter � 2008 Elsevier Ltd. A doi:10.1016/j.eswa.2008.05.049 * Corresponding author. Tel.: +886 3 5726731; fax: E-mail address: shhsu@cc.nctu.edu.tw (S.H. Hsu). a b s t r a c t Roles in video games often serve as avatars of players. Different game players may have their particular preferences on a role’s facial appearance. It would be desirable to allow players to customize the design of roles. This paper presents two methods for recommending a roles’ facial appearance for a particular game player and illustrates the two methods by using heroic roles as an example. The two recommendation methods are designated as the text-input and the picture-input approaches. The text-input approach requests the game player to carry out pairwise comparisons for determining the relative weights of 16 personality traits of heroes. The recommendation mechanism for the text-input approach is based on the fuzzy AHP (analytic hierarchy process). Whereas the picture-input approach requests the game player to view a sample set of pictures and rate his/her preferences on each picture. The recommendation mech- anism for the picture-input approach is based on the BP (back-propagation) neural network. Experiments indicated that the text-input approach is more effective in terms of recommending an appropriate facial appearance, yet at the expense of needing more user time. � 2008 Elsevier Ltd. All rights reserved. 1. Introduction In playing a video game (called a game hereafter), a player ex- presses his or her intentions by manipulating the actions of a role. A game role is a character, which serves as the player’s avatar or competitor. Roles to a game are as important as actors to a film. Casting appropriate actors can lead to the success of a film. In the same way, designing appropriate roles is very important to the success of a video game. Previous researchers have published numerous studies on the role design of video games. Most of them attempted to create a life-like role by emulating human behaviors—such as dialogue (Brusk & Eladhari, 2006; Gustafson, Boye, Fredriksson, Johanneson, & Königsmann, 2005; Jan & Traum, 2005), intelligence (Frasca, 2001; Lair & Duchi, 2000; Vala, Paiva, & Prada, 2004), motor-skill actions (Blumberg & Galyean, 1997), and emotional expressions (Rizzo, Neumann, Enciso, Fidaleo, & Noh, 2001; Wallraven, Breidt, Cunningham, & Bülthoff, 2005). Of these studies, those which analyze emotional expressions attempt to automatically create a facial expression to represent a character’s mood (e.g. happy, angry, sad, and disgusted). Such fa- cial expression studies have an implicit objective—mood manifesta- tion (Bartlett, Hager, Ekman, & Sejnowski, 1999; Ekman, 1993; Zhang & Ji, 2005). That is, a computer-generated facial expression should model a particular mood (e.g. sad) that is easily recogniz- able by humans. ll rights reserved. +886 3 5722392. In a film, proper recognition of an actors’ mood by interpreting their body languages is never enough. To produce a popular film, the attractive appearance of actors is often much more important. Likewise, the appearance of a game role may influence player involvement in the game. An attractive appearance may induce players to have affectionate feelings for the avatar, and in turn, make game play more fun (Hsu, Lee, & Wu, 2005). Even though the facial appearance of a role is very important to game design, this topic has rarely been investigated. This study proposes two methods for automatically recom- mending attractive facial appearances of heroic roles to game play- ers—in a customization manner. The observations in this study indicate that different game players have different preferences for the appearance of a hero. According to further interviews, this difference is due to the fact that players have different preferences for a hero’s personality traits. This implies that personality traits may be manifested by facial appearances. Based on this implication, this study proposes a research frame- work to design a hero’s facial appearance as favored by a particular game player (Fig. 1). This research framework involves two phases: creation and application. The creation phase develops a database that relates facial appearance to personality traits, which involves two steps. Firstly, multimedia software is used to create samples of her- oes’ facial appearances. Secondly, game players evaluate the personal- ity traits for each facial appearance. In this phase, a vector comprising numeric data can model both a facial appearance and its personality traits. The created database is called a face-trait database. The application phase implements the customization design of the heroes’ facial appearance. That is, for any given game player, mailto:shhsu@cc.nctu.edu.tw http://www.sciencedirect.com/science/journal/09574174 http://www.elsevier.com/locate/eswa Text-input recommendation mechanism (Fuzzy AHP) Picture-input recommendation mechanism (Neural network) Establish face-trait database Creation phase Application phase Fig. 1. Research framework. 4930 S.H. Hsu et al. / Expert Systems with Applications 36 (2009) 4929–4934 a highly favored heroes’ facial appearance can be quickly recom- mended from the face-trait database. Two recommendation mech- anisms can be used to retrieve these highly favored facial appearance profiles. One mechanism is a text-input approach, which requires the game player to determine each personality trait’s relative weight. This mechanism uses the fuzzy AHP technique (Laarhoven & Ped- rycz, 1983; Saaty, 1980; Shamsuzzaman, 2000; Zadeh, 1975). The other mechanism is a picture-input approach, which requires the game player to evaluate their degree of preferences for a sample set of pictures (facial appearances). This mechanism is imple- mented using the back-propagation (BP) neural network technique. The remainder of this paper is organized as follows. Section 2 describes how the face-trait database is created. Section 3 first introduces the fuzzy AHP technique, and then presents the text- input recommendation mechanism. Section 4 describes the picture-input recommendation mechanism. Section 5 presents the experiments and results for justifying the effectiveness of the two recommendation mechanisms. The last section contains concluding remarks. Fig. 2. The facial appearance creation so 2. Face-trait database A prototype face-trait database has been developed. In develop- ing the database, heroes’ facial appearances were created by com- mercially available multimedia software (Live Studio Head Tool V2.6). We use 16 personality traits of heroes to characterize each fa- cial appearance created. A hero’s facial appearance could then be encoded by a vector consisting of 16 elements, which provides a key for the recommendation facial appearances. 2.1. Facial appearance creation The multimedia software used to create a facial appearance is based on a feature-based approach. That is, the configuration of a face is composed of several features. A feature may denote a part of a face such as eyes, lips, and nose or denote a face’s characteristic such as skin color and texture. For each feature, there are many op- tions for selection. Various combinations of feature options result in different facial appearances. For features associated with geometric shapes, their feature op- tions are created by the parametric-geometry approach (Myung & Han, 2001; Verroust, Schonek, & Roller, 1992). That is, changing some of its geometric parameters can vary the shape of an object. Considering an object with rectangular shape as an example, we could take the length-to-width ratio as one parameter and vary shape of the object by changing the parameter value. Likewise, a facial feature such as eyes can be modeled by a parametric-geom- etry shape with more than one parameter; for example, the curva- ture of upper eye lid and that of lower eye lid. By varying these two parameters, we could have many different shapes for modeling eyes. Example features associated with geometric shapes include lips, noses, eyes, eyebrows, and facial outline. These shapes are com- monly controlled by a set of parameters, and each parameter value ftware Live Studio Head Tool v.2.6. S.H. Hsu et al. / Expert Systems with Applications 36 (2009) 4929–4934 4931 is a real number in a predefined range such as [0, 1]. The input of such a real-number value is through a scale-bar input device (rightmost part of Fig. 2). Due to the completeness property of real numbers, there are theoretically an infinite number of options available for features associated with geometric types. By contrast, some other features provide only a finite number of feature options; for example—type of hair and type of moustache. The input of such a nominal-type feature option is through the clicking of a menu box (middle part of Fig. 2). The multimedia software we used provides 12 features. In this research, only five shape-oriented features are chosen as variables in designing a facial appearance for the prototype database. These five features are facial outline, eye brows, eyes, lip, and nose, which are selected because most prior research has concluded their importance on facial expression and recognition (Adolphs, 2002; Brunelli & Poggio, 1993; Sadrô, Jarudi, & Sinha, 2003). The prototype database which is essentially expandable, at its present state, includes 243 different facial appearances. That is, each of the five facial features has only three options for selection, which leads to 35 = 243 different facial appearances. Three in- stances of the facial appearances are shown in Fig. 3. 2.2. Personality traits evaluation We use 16 personality traits to characterize each facial appear- ance. These personality traits are a hero’s characters reported in a prior research by Hsu, Kao, and Wu (2007). According to their re- search, by the technique of factor analysis (principal components factoring) (McDonald, 1985), these 16 personality traits can be cat- egorized into three groups—bravery, visionary, and moral as shown in Fig. 4. In the characterization of a facial appearance, a five-point scale evaluates each personality trait. The higher the value, the higher degree does the facial appearance reveal—in terms of the personal- ity trait. In the characterization process, we asked 112 subjects (61males, 51 females at the age of 17–25) to evaluate the 16 per- Fig. 3. Example of facial appearances. He Bravery Visio C ourageous U nbeatable C harism atic M ighty C ool A m bitious P ersevering V isionary D aring Fig. 4. Three levels fuzzy A sonality traits for each of 243 facial appearances. All these subjects are all game player, who are either senior high school or college students. Results of the characterization process yield 243 trait vectors, each of which represents a particular facial appearance. The value of each element, ranging from 1 to 5, is the mean score reported by all the subjects. Let X = [x1, . . ., x16] denote a trait vector so obtained, which is further transformed into a normalized vector, denoted by Y ¼ ½y1; . . . ; y16� where yi ¼ xiP16 k¼1 xk and P16 i¼1 yi ¼ 1. 3. Text-input recommendation mechanism To create a hero face favored by a particular game player, it is important to know how the player weights each personality trait. The procedure for determining the weights of personality traits is by applying the fuzzy AHP technique (Saaty, 1990). The technique, widely applied in various areas (Durán & Aguilo, 2007; Kim & Yoon, 1992; Mamaghani, 2002; Muralidar & Santhanam, 1990; Wu, Lo, & Hsu, 2007) is briefly stated below. Firstly, the 16 personality traits are hierarchically clustered as shown in Fig. 4. This hierarchy indicates that these 16 personality traits, based on a prior research (Hsu et al., 2007), can be catego- rized into three groups—bravery, visionary, and moral, which are called group-level traits. Each personality traits within a group is called a member-level trait. Secondly, the relative weights for group-level traits and that for member-level traits in each group have to be determined. We there- fore have four weight-determination problems, one for group-level and three for member-level traits. To each weight-determination problem, we asked the game player to carry out a pairwise com- parison experiment, and use a fuzzy AHP algorithm to process the experiment data for determining the relative weights. Thirdly, we use the relative weights to compute a recommenda- tion-priority value to retrieve a hero face from the face-trait data- base for the game player. 3.1. Pairwise comparison experiment The pairwise comparison experiment is explained by using the weight-determination of group-level traits as an example. Of the three group-level traits, any two (a pair) is chosen for comparison. Assume the pair (bravery, visionary) is chosen. Then, the player is asked to answer the following question: ‘‘Consider a person to be recognized as a hero. Of the two groups of personality traits, which one is more important? Try to compare the degree of importance between them.” Now suppose the player answers bravery is more important than visionary. Then, five linguistic variables listed in Table 1 are provided for the player to express his/her opinion. These linguistic variables range from ‘‘absolutely important” to ‘‘equally impor- tant” on a five level scales, where between any two consecutive ro nary Moral L eadership C apable T rustw orthy C alm G racious Just U nselfish HP hierarchy model. Table 1 Linguistic variables used in the fuzzy AHP Fuzzy number Linguistic variables ~1 ¼ð1; 1; 1Þ Equally important ~3 ¼ð2; 3; 4Þ Weakly important ~5 ¼ð4; 5; 6Þ Essentially important ~7 ¼ð6; 7; 8Þ Very strongly important ~9 ¼ð8; 9; 10Þ Absolutely important ~2 ¼ð1; 2; 3Þ; ~4 ¼ð3; 4; 5Þ; ~6 ¼ð5; 6; 7Þ; ~8 ¼ð7; 8; 9Þ Intermediate values between two adjacent judgments 4932 S.H. Hsu et al. / Expert Systems with Applications 36 (2009) 4929–4934 scales an intermediate scale is additionally defined so that nine scales are finally created. As shown in Table 1, each linguistic var- iable is represented by a triangular fuzzy number—for example, ~7 ¼ð6; 7; 8Þ denotes ‘‘very strongly important”. The adoption of lin- guistic variables is to resolve the vagueness occurred in human judgment (Zadeh, 1975). Based on the pairwise comparison experiment, a n � n matrix à = [ãij] could be obtained, where n denotes the number of traits to be compared, ~A ¼ 1 ~a12 � � � ~a1n ~a21 1 � � � ~a2n .. . .. . . . . .. . ~an1 ~an2 � � � 1 2 66664 3 77775 Notice that ãij = 1/ãji, which ensures that the comparison for each pair (i, j) is consistent; and ãii = 1, which denotes that a self-compari- son is always ‘‘equally important” and is not needed. 3.2. Fuzzy AHP algorithm Two procedures are used to obtain the relative weights for group-level and member-level traits. The first one Compute_Rela- tive_Weight is intended to compute the relative weights, and the second one Validity_Check_for_Relative_Weights is developed for checking the validity of the obtained relative weights. These two procedures are respectively described below, where the definitions of arithmetic operators (i.e., �, �, and Defuzzy) are introduced in the Appendix. 3.2.1. Procedure Compute_Relative_Weight Step 1: Calculate ~W i , the fuzzy weight for each row i in à (Buckley, 1985) ~Zi ¼ð~ai 1 � ~ai 2 ����� ~ainÞ 1 n; 8i ¼ 1; 2; . . . ; n ~W i ¼ ~Zi �ð~Z1 � ~Z2 ��� �� ~ZnÞ �1 ; 8i ¼ 1; :::; n Step 2: Defuzzication of ~W i and à (Teng & Tzeng, 1993) Ŵ i ¼ Defuzzyð ~W iÞ aij ¼ Defuzzyð~aijÞ; whereA ¼ ½aij� and ~A ¼ ½~aij� Step 3: Compute relative weights W_{i} W i ¼ Ŵ iPn i¼1 Ŵ i Table 2 Values of RI n 1 2 3 4 5 6 7 8 9 10 11 RI 0.00 0.00 0.58 0.90 1.12 1.24 1.32 1.41 1.45 1.49 1.51 3.2.2. Procedure Validity_Check_for_Relative_Weights Step 1: Compute W�i W�1 W�2 .. . W�n 2 66664 3 77775 ¼ A W 1 W 2 .. . W n 2 66664 3 77775 Step 2: Compute kmax, the maximum eigenvalue kmax ¼ 1 n W�1 W 1 � � þ W�2 W 2 � � þ�� �þ W�n W n � �� � Step 3: Compute CI, the consistency index CI ¼ kmax � n n � 1 Step 4: Compute CR CR = CI/RI the values of RI are shown in Table 2 (Saaty, 1980). Step 5: Consistency check If CR 6 0.1 (pairwise comparison data à is reasonably consistent) Output the resulting relative weights Wi (1 6 i 6 n) Else (CR > 0.1 pairwise comparison data à are inconsistent) Repeat the pairwise comparison experiment. Endif 3.3. Recommending mechanism To recommend a hero face favored by a particular game player, we first compute a recommendation-priority value for each facial appearance in the face-trait database and recommend the one with the highest recommendation-priority value. Define S = {(Yi, Fi)|1 6 i 6 n} as the face-trait database developed for facial design, where Fi denotes ith facial appearance, and Yi = [yij], 1 6 j 6 16 denotes the personality-trait vector of Fi Notice that Fi is a 2D picture while Yi is a vector with 16 numeric elements, where yij denotes the degree of jth personality traits that picture Fi manifest to a ‘‘common people”. Let W = [wj], 1 6 j 6 16 represent the preferences of a particular game player on each of the 16 per- sonality traits, where wj denotes the relative weight of jth person- ality trait, perceived by the game player—an ‘‘individual people” rather than a ‘‘common people”. The procedure for recommending facial appearances most favorable to a particular game player proceeds as follows. Step 1: Compute the recommendation-priority value pi ¼ X16 j¼1 wj � yij; 1 6 i 6 n Step 2: Output the recommended one i� ¼ arg maxðpiÞ 4. Picture-input recommendation mechanism The BP neural network technique has been widely used as a pre- dictor (Dutta & Shekhar, 1988; Liau & Chen, 2005; O’Leary, 1998; Salchenberger, Cinar, & Lash, 1992; Tam & Kiang, 1992). Given a sample set of input/output data obtained from a real-world sys- tem, we can use the technique to establish a BP network that serves as an input/output mapping mechanism—a three-layer architecture as shown in Fig. 5. The established BP network can be used to predict the output for a new input data set. Details of the BP neural network technique can be referred to Wasserman (1989) and Hertz, Krogh, and Palmer (1991). The picture-input recommendation mechanism is to establish a BP network for a particular game player. For such a BP network, its Input Hidden Output fi1 fi2 fi3 fi4 fi15 Pi Fig. 5. Architecture of the neural network. S.H. Hsu et al. / Expert Systems with Applications 36 (2009) 4929–4934 4933 input is a facial appearance denoted by Fi = [fik], 1 6 i 6 243, 1 6 k 6 15, where fik (a binary number) denotes the existence of the kth feature option in the ith facial appearance. As stated, a facial appearance has five features, each of which has three options for selection. This leads to 15 (5 � 3) types of feature options. A facial appearance can then be modeled by a vector with 15 binary ele- ments, in which 1 denotes a feature option exists and 0 denotes its nonexistence. The output of such a BP network is denoted by Pi, which denotes the game player’s preference on the ith facial appearance. The use of the BP neural network technique involves two phases: training and predicting phases. In the training phase, we at- tempt to establish a BP network for the game player. Out of the 243 facial appearances, we randomly sample 81 ones and use them to build up a BP network. The algorithms of the training process can be referred to Grossberg (1974) and Rumelhart, Hinton, and Wil- liams (1986). In the predicting phase, we attempt to use the estab- lished BP network to predict the player’s preferences on the other facial appearances not considered in training phase. That is, the BP network will predict the player’s preference on the remaining 162 facial appearances. 5. Experiments An experiment is carried out to compare the performance of the text-input and the picture-input approaches. Twenty game players (10 males, 10 females at the age of 17–25) are invited as experi- ment subjects, and 243 facial appearances are created in the face-trait database. The experiment proceeds as follows. Firstly, we carried out the text-input approach in which each subject i is requested to perform a pairwise comparison for deter- mining his/her relative weights on the 16 personality traits. The system will output a picture (say, X�i ) with the highest recommen- dation-priority value. Secondly, we carried out the picture-input approach in which each subject i is requested to view 81 pictures and give his/her preference on each picture. The input/output data of the 81 pic- tures are used to establish a BP network. Then, all the 243 pictures are fed into the BP network, and the one (say, Y�i ) with the highest preference value will be recommended. Table 3 Comparing effectiveness Subjects 1 2 3 4 5 6 7 8 9 10 Text-input xi 0 0 0 0 0 1 0 1 0 0 Picture-input yi 0 0 3 0 1 2 0 2 0 1 Finally, each subject is requested to view the remaining 162 pic- tures and give his/her preference on each one. Out of the 243 pic- tures, the one with the highest preference (say, Z�i ) is selected. Define rðP�i Þ as the preference of picture P evaluated by subject i. The effectiveness of the text-input approach is measured by the distribution of the metric: xi ¼ rðZ�i Þ� rðX � i Þ, and that for the pic- ture-input approach is measured by the distribution of yi ¼ rðZ�i Þ� rðY � i Þ . The distributions of xi and yi are shown in Table 3. The table indicates that the text-input approach is superior to the picture-in- put approach. In the text-input approach, 85% game players (17 out of the 20 subjects) will get their most favor picture, while only 35% can get so in the picture-input approach. However, the text-input approach needs more user time. The average time for performing a pairwise comparison is about 26 min. and that for evaluating 81 pictures is about 2.6 min. The experiment results lead to the following implication. For a video game equipped with relatively few numbers of pictures, we would suggest an exhaustive display of pictures to a game player. In contrast, for a video game equipped with a great amount of pictures, we would suggest the use of the text-input approach. 6. Conclusions Playing video games through manipulating a role is quite com- mon. Role design therefore has been a significant research issue. Most prior research attempted to create a life-like role by develop software for emulating human’s capabilities, such as dialogues, intelligence, motor-skill, and emotional expressions. This research is unique in providing a customized facial appearance for each game player. A video game with such a customization function would become more popular. Two methods for providing such a customization function have been developed. One is called the text-input approach whose mechanism is based on the fuzzy AHP technique. The other is called the picture-input approach whose mechanism is based on the BP neural network technique. The text-input approach requires a game player to perform a pairwise comparison on 16 personality traits. The picture-input approach requires a game player to view and evaluate a sample set of pictures. Experiment results indicated that the text-input approach is superior to the picture-input approach, in terms of recommending an appropriate picture to a game player; however, at the price of needing more user input time. Appendix. Arithmetical operators for fuzzy numbers The fuzzy arithmetic operators used in this research are defined below (Laarhoven & Pedrycz, 1983; Zadeh, 1975), by referring two fuzzy numbers ã1 = (l1, m1, r1) and ã2 = (l1, m1, r1). (1) Addition operator: �ã1 � ã2 = (l1 + l2, m1 + m2, r1 + r2) (2) Multiplication operator: �ã1 � ã2 = (l1 � l2, m1 � m2, r1 � r2) (3) Defuzzication operator (Teng & Tzeng, 1993) Defuzzy(ã1) = |(r1 � l1) + (m1 � l1)|/3 + l1 11 12 13 14 15 16 17 18 19 20 0 0 0 0 0 0 1 0 0 0 0 1 1 2 0 2 1 1 3 1 4934 S.H. Hsu et al. / Expert Systems with Applications 36 (2009) 4929–4934 References Adolphs, R. (2002). Recognizing emotions from facial expressions: Psychological and neurological mechanisms. Behavioral and Cognitive Neuroscience Reviews 1, 1, 21–61. Bartlett, M. S., Hager, J. C., Ekman, P., & Sejnowski, T. J. (1999). Measuring facial expressions by computer image analysis. Psychophysiology, 36, 253–263. Blumberg, B., & Galyean, T. (1997). Multi-level control for animated autonomous agents: Do the right thing. . . oh, not that. . .. In: R. Trappl & P. Petta (Eds.), Creating personalities for synthetic actors. Berlin: Springer-Verlag. (pp. 74–82). Brunelli, R., & Poggio, T. (1993). Face recognition features versus templates. IEEE Transactions on PAMI, 15(10), 1042–1052. Brusk, J., & Eladhari, M. (2006). Playing the character. RPG seminar, Tampere, March (pp. 30–31). Buckley, J. J. (1985). Ranking alternatives using fuzzy numbers. Fuzzy Sets and Systems, 15(1), 21–31. Durán, O., & Aguilo, J. (2007). Computer-aided machine-tool selection based on a Fuzzy-AHP approach. Expert Systems with Applications. doi:10.1016/ j.eswa.2007.01.046. Dutta, S., & Shekhar, S. (1988). Bond rating: A non-conservative application of neural networks. In Proceedings of the IEEE international conference on neural networks (pp. 443–450). Ekman, P. (1993). Facial expressions and emotion. American Psychologist, 48, 384–392. Frasca, G. (2001). Rethinking agency and immersion: Videogames as a means of consciousness-raisin (essay). SIGGRAPH 2001 N-Space art gallery. Grossberg, S. (1974). Classical and instrumental learning by neural networks. Progress in theoretical biology (Vol. 3, pp. 51–141). New York, NY: Academic Press. Gustafson, J., Boye, J., Fredriksson, M., Johanneson, L., & Königsmann, J. (2005). Providing computer game characters with conversational abilities. In Proceedings of intelligent virtual agent (IVA05), Kos, Greece. Hertz, J., Krogh, A., & Palmer, R. G. (1991). Introduction to the theory of neural computation. Redwood City, CA:: Addison-Wesley. Hsu, S. H., Kao, C. H., & Wu, M. C. (2007). Factors influencing player preferences for heroic roles in role-playing games. CyberPsychology & Behavior, 10(2), 293–295. Hsu, S. H., Lee, F. L., & Wu, M. C. (2005). Designing action games for appealing to buyers. CyberPsychology & Behavior, 8(6), 585–591. Jan, D., Traum, D. R. (2005). Dialog simulation for background characters. In IVA (pp. 65–74). Kim, C. S., & Yoon, Y. (1992). Selection of a good expert system shell for instructional purposes in business. Information & Management, 23(5), 249–262. Laarhoven, P. J. M., & Pedrycz, W. (1983). A fuzzy extension of Saaty’s priority theory. Fuzzy Sets and Systems, 11(3), 229–241. Lair, J. E., & Duchi, J. C. (2000). Creating human-like synthetic characters with multiple skill levels: A case study using the Soar Quakebot. AAAI 2000 fall symposium on simulating human agent, technical report FS-00-03 (pp. 75–79). AAAI Press. Liau, L. C. K., & Chen, B. S. C. (2005). Process optimization of gold stud bump manufacturing using artificial neural networks. Expert Systems with Applications, 29, 264–271. Mamaghani, F. (2002). Evaluation and selection of an antivirus and content filtering software. Information Management & Computer Security, 10(1), 28–32. McDonald, R. (1985). Factor analysis and related techniques. Hillsdale, NJ: Lawrence Erlbaum. Muralidar, K., & Santhanam, R. (1990). Using the analytic hierarchy process for information system project selection. Information & Management, 18(1), 87–95. Myung, S., & Han, S. (2001). Knowledge-base parametric design of mechanical products based on configuration design method. Expert Systems with Applications, 21, 99–107. O’Leary, D. E. (1998). Using neural networks to predict corporate failure. International Journal of Intelligent Systems in Accounting, Finance and Management, 7, 187–197. Rizzo, A. A., Neumann, U., Enciso, R., Fidaleo, D., & Noh, J. Y. (2001). Performance- driven facial animation: Basic research on human judgments of emotional state in facial avatars. CyberPsychology & Behavior, 4(4), 471–487. Rumelhart, D. E., Hinton, G. E., & Williams, R. J. (1986). Learning internal representation by error propagation. Parallel and Distributed Processing, 1, 318–362. Saaty, T. L. (1980). The analytic hierarchy process: Planning, priority setting, resource allocation. New York: McGraw-Hill (p. 20). Saaty, T. L. (1990). How to make a decision: The analytic hierarchy process. European Journal of Operational Research, 48(1), 9–26. Sadrô, J., Jarudi, I., & Sinha, P. (2003). The role of eyebrows in face recognition. Perception, 32, 285–293. Salchenberger, L. M., Cinar, E. M., & Lash, N. A. (1992). Neural networks: A new tool for predicting thrift failures. Decision Sciences, 23(4), 899–916. Shamsuzzaman, M. (2000). Selection of a FMS based on fuzzy set theory and AHP methods. ISE-Thesis, Asian Institute of Technology, Bangkok. Tam, K. Y., & Kiang, M. Y. (1992). Managerial applications of neural networks: The case of bank failure predictions. Management Science, 38(7), 926–947. Teng, J. Y., & Tzeng, G. H. (1993). Transportation investment project selection with fuzzy multi-objective. Transportation Planning and Technology, 17, 91–112. Vala, M., Paiva, A., & Prada, R. (2004). From motion control to emotion influence: Controlling autonomous synthetic characters in a computer game. In AAMAS’04, New York, USA (pp. 1300–1301). Verroust, A., Schonek, F., & Roller, D. (1992). Rule-oriented method for parameterized computer-aided design. Computer Aided Design, 24(10), 531– 540. Wallraven, C., Breidt, M., Cunningham, D. W., & Bülthoff, H. H. (2005). Psychophysical evaluation of animated facial expressions. In Proceedings of the 2nd symposium on Applied perception in graphics and visualization (pp. 26–28). Wasserman, R. (1989). Neural computing: Theory and practice. New York, NY: Van Nostrand Reinhold. Wu, M. C., Lo, Y. F., & Hsu, S. H. (2007). A fuzzy CBR technique for generating product ideas. Expert Systems with Application, 34, 530–540. Zadeh, L. A. (1975). The concept of a linguistic variable and its application to approximate reasoning. Information Sciences, Part 1, 8, 199–249; Part 2, 8, pp. 301–357; Part 3, pp. 43–80. Zhang, Y., & Ji, Q. (2005). Active and dynamic information fusion for facial expression understanding from image sequences. IEEE Transactions on Pattern Analysis and Machine Intelligence, 27(5), 699–714. http://dx.doi.org/10.1016/j.eswa.2007.01.046 http://dx.doi.org/10.1016/j.eswa.2007.01.046 Design Facial Appearance facial appearance for Roles roles in Video Gamesvideo games Introduction Face-trait database Facial appearance creation Personality traits evaluation Text-input recommendation mechanism Pairwise comparison experiment Fuzzy AHP algorithm Procedure Compute_Relative_Weight Procedure Validity_Check_for_Relative_Weights Recommending mechanism Picture-input recommendation mechanism Experiments Conclusions Arithmetical operators for fuzzy numbers References