2012 45h Hawaii Inernaional Conference on Sysem Sciences Building Social Decision Suppor Mechanisms wih Friend Neworks Yung-Ming Li Insiue of Informaion Managemen, Naional Chiao Tung Universiy, Hsinchu, Taiwan 300, ROC yml@mail.ncu.edu.w Yi-Lin Lee Insiue of Informaion Managemen, Naional Chiao Tung Universiy, Hsinchu, Taiwan 300, ROC marinerlee@gmail.com Absrac Many Inerne users are geing used o make decisions based on opinions colleced from heir own social neworks. While convenional decision suppor sysem has been exensively invesigaed, lile specific mechanism, however, on how social neworks can help users wih decision-making is developed. Using social nework analysis along wih regression model, fuzzy Delphi and fuzzy AHP mehods as ools, his paper designs a social nework based decision suppor sysem wih beer effeciveness. 1. Inroducion Social Neworking Sies (SNS) are increasingly popular nowadays. Sies like Facebook and Twier are compeing o give he bes feaures in erms of socializaion, ineracion and even enerainmen. However, he main purpose of using social neworks should be socializing wih people who you like and are ineresed in. I is no surprising ha many people are hooked on wih hese sies, bu predicions abou he fuure of social neworks are various. 1.1. Curren applicaions of social nework As social neworking echnology evolves and increasing online aciviies make more daa ses available, social neworks are beginning o be uilized in many ways. Mos social neworking services are used for sharing wha you ve done or wha you re doing, bu he applicaions should no be focused on personal aciviies only. Curren developmen of social nework analysis (SNA) is limied o discovering friend connecions, leaders, influenial people and friends. For example, markeing applicaions have been widely discussed o discover poenial opporuniies for aracing cusomers, bu he facors considered in hese applicaions are limied o personal connecions. Human decision making is ofen based on heurisics han on pure logic or weighing advanages and disadvanages. When people make decisions hey are ofen influenced by wha oher people like hemselves are saying and doing. Missing from social nework sudies, however, has been discussions of decision mechanisms. The social nework approach focuses on relaions beween people, raher han on aribues of people. For example, i measures he exisence of friendship beween individuals, raher han he friendliness of each individual [16]. In moving from he analysis of aribues oward relaion-cenered analysis, nework researchers appeared o have los sigh of ime and is poenial effecs on he srengh of relaions. Wha new insigh could we discover if social nework analysis akes ime ino accoun, in addiion o only connecions beween people? If new facors could be aken ino consideraion, specifically changes over ime, hen social nework analysis could discover hings like emergence of leadership and changes in rus over ime. And ha will be very valuable informaion ha he social neworks have barely begun o capure. 1.2. Saemen of he research problem Wih respec o decision making in groups, wo aspecs are imporan [7]: normaive and informaional influence. The former one is based on he desire o conform o he expecaions of ohers, and he laer one is based on he accepance of informaion from ohers [14]. Recenly, some simple decision suppor funcions like Quesions on Facebook have been developed (Figure 1). By using Quesions you can ge recommendaions or conduc polls. However, i is no a real decision suppor ool because only voing funcion is provided. Besides, he crieria have o be seup before he quesion is presened o friends. Tha is, he 978-0-7695-4525-7/12 $26.00 2012 IEEE DOI 10.1109/HICSS.2012.135 1777
decision maker should have clear undersanding abou quesions, and have he abiliy o lis decision crieria. Typically, we may ask friends quesions like: Wha o buy? You may ge a lis of recommended producs from recommendaion mechanism like conen-based or collaboraive filering, bu lile informaion abou how he producs are seleced is known. If users ry o make decision based on funcions like Quesion, hey should have clear undersanding abou decision problems and have o be capable of defining decision crieria. However, if his is no he case, exising funcions won work well. This aricle aemps wo asks. Firs, he main obecive is o design a social decision suppor sysem running on online social nework. By disclosing he power of online social nework, we inroduce ime facor o expand radiional social nework analysis, in which he inerconnecions are core. Second, an improved similariy aggregaion mehod was proposed o beer fi he characerisics of social nework. We inend o answer quesions like: How o buy? Tha is, we ake a sep forward by supporing decision makers wih crieria discovering, decision group selecing and decision crieria synhesizing. A decision maker does no o be familiar wih decision quesions, nor does he ake care of decision processes. When a problem is issued, our sysem help users wih selecing decision group members from users social nework, collecing opinions and aggregaing hem ino alernaives. Users can undersand he whole picure of wha he alernaives are based on. on Facebook. Third, an adapive fuzzy similariy aggregaion mehod was also presened o beer fi he characerisics of informaional influence wihin social neworks. The remaining par of his paper is organized as follows. In secion 2, we discuss exising lieraure relaed o our research opics. Secion 3 demonsraes our research model and in secion 4 we describe he experimenal daa source, seings, and procedures. The experimenal resuls and evaluaions are discussed in secion 5. Finally, secion 6 concludes our research conribuions and presens fuure research direcions. 2. Lieraure review 2.1. Social decision suppor Social decision making is he process ha akes every individual s local decisions and generaes a collecive response [25]. Social decision suppor (SDS) heory is composed of individual preferences, group composiion, group influence processes, and collecive responses [26]. I has also been applied o problem-solving asks [19] and collecive recall [10]. A social decision suppor sysem (SDSS) allows users in a nework-based environmen o form a decision group and paricipae in a collaboraive decision making process. The social decision suppor sysem (SDSS) idea was firs inroduced in [29] as a compuer suppored group decision making sysem. A social decision suppor sysem draws is foundaion from a number of relaed areas such as Delphi mehod [20], compuer mediaed communicaion [9], decision heory [2, 8] and online communiies [24]. However, we hink ha he design of sysems o suppor online socieal collaboraive processes for formulaion of decision crieria and alernaives requires some new and challenging developmen for successful operaion. 2.2. Social nework analysis Figure 1. Quesions funcion on Facebook 1.3. Significance of he sudy In our sudy, we seleced famous social nework sie Facebook o be our experimen plaform. Our sudy differeniaes from exising works in hree ways. Firs, we measured he friendship by combining ime facor wih cenraliy from social nework analysis. Second, a regression model was used o build up decision group selecing mechanism based on he friend relaionships Social nework analysis is becoming increasingly popular as a general mehodology for undersanding complex paerns of ineracion. I provides a visual represenaion of users and inerconnecions. The mos popular merics used are degree, beweenness and closeness cenraliy [5]. A user wih higher beweenness cenraliy is ofen considered as an opinion leader [5], and a higher closeness cenraliy indicaes ha a user is highly relaed o all ohers [23]. The sudies of social neworks have examined a diverse se of properies, and hese properies are classified as relaional properies and srucural properies [27]. Relaional properies focus on he 1778
conen of he relaionship beween nework members and on he form of hese relaionships, while srucural properies describe he way members fi ogeher o form social neworks. Human relaionships are mainained, are renewed, or deeriorae over ime [28], bu ime facor is missing from he above properies. 2.3. Fuzzy Delphi and Fuzzy AHP Delphi mehod is a ype of collecive decision-making mehod [20, 33], and i requires several rounds of anonymous wrien quesionnaire surveys for he purpose of collecing expers opinion. This mus coninue asking expers for heir opinion unil he expers arrive a a consensus. As a resul, i generally has weaknesses such as ime and cos consuming. Fuzzy Delphi Mehod (FDM) is an analyical mehod based on he Delphi Mehod ha draws on he ideas of he Fuzzy Theory. To solve he problem of fuzziness in exper consensus in group decision making, researchers from around he world came up wih new mehods [15, 22]. Combining wih cumulaive frequency and fuzzy scoring, Ishikawa e al. [13] proposed he Maximum-Minimum Mehod o compile he expers opinions ino fuzzy numbers. Hsu and Chen [11] proposed a fuzzy similariy aggregaion mehod, in which similariies beween expers were collaed and fuzzy numbers assigned direcly o each exper o deermine he agreemen degree beween hem. The consensus coefficien was hen used o aggregae all expers fuzzy evaluaion values. Analyic Hierarchy Process (AHP) is a srucured, muli-crieria decision-making approach for dealing wih complex, ill-srucured problems by organizing he decision facors in a hierarchical srucure. I is one of he mos popular and powerful mehods for group decision making. Besides, AHP is widely used for dealing wih quanifiable and inangible crieria ha can be applied o numerous areas such as decision heory [31]. While AHP can be used o capure he prioriies of individual decision paricipans, i is necessary o combine hese individual assessmens ino a consensus. Though he purpose of convenional AHP is o capure he expers knowledge, i sill canno reflec he human hinking syle. Therefore, Laarhoven and Pedrycz [18] proposed a fuzzy exension of AHP o solve his issue. The linguisic scale of radiional AHP mehod could express he fuzzy uncerainy when a decision maker is making a decision. Therefore, FAHP ranslaes he opinions of expers from definie values ino fuzzy numbers and membership funcions, presens riangular fuzzy numbers in paired comparison of marices o develop FAHP. Consequenly, he opinions of expers approach human hinking model, so as o achieve more reasonable evaluaion crieria. 3. Sysem framework Our requiremens for his sysem are governed by he obecive of designing a sysem o suppor decision processes on social nework. Typically, a group decision process includes choosing he expers, deermining he evaluaion crieria, aggregaing expers crieria and suggesing alernaives. For more vivid picure of he sudy, Figure 2 served as he research paradigm. In he following, we describe our imporan sysem modules in deail. Figure 2. Sysem flow 3.1. Expers discovering People's brains are more responsive o friends han o srangers, even if he sranger has more in common [17]. There are psychological and evoluionary argumens for he idea ha he social facors of similariy and closeness could ge privileged reamen in he brain. However, a sudy suggess ha social closeness is he primary facor, raher han social similariy, as previously assumed [17]. As SNA is inroduced o analyze complex neworks [3], in our model we choose closeness from hree commonly used cenraliy merics o be one of our firs exper selecion facors. Closeness cenraliy is defined as he oal disance of a user from all oher users, and can be formulaed as [3]: n C ( p ) = 1/ d p, p (1) ( ) c i i = 1 where n is he number of users and d( pi, p) is he disance beween decision maker i and his friend. Beweenness cenraliy racks he number of geodesic pahs hrough he enire social nework, which pass hrough he concep whose influence is measured. I is an approximaion of is influence on he 1779
discussion in general [3]. Besides, beweenness cenraliy bes measures which members, in a se of members, are viewed mos frequenly as a leader, han oher social nework analysis measures [5]. The beweenness cenraliy evaluaes he capabiliy of ineracions beween wo friends and is defined as [6]: C () i = ( g () i )/ G (2) B l l i l where G l is he number of he shores pahs linking wo friends (, i ) and g l () i is he number of shores pahs linking he wo nodes ( l, ) conaining node i. However, even wo people are close friends; friendship may evaporae as ime goes by if hey do no inerac frequenly. To measure how he friendship changes wihin a ime period, we define an evaporaion funcion o be a new facor of evaluaing friendship among friends. The evaporaion funcion is formulaed as: 1 τ + i = (1 ρi ) τi + τi (3) where: τ : friendship deposied for friend i ρ i : friendship evaporaion coefficien, L i L ρi = (4) S N τ i : amoun of friendship changed in ime, / if ineracion exiss beween (, ) Q Li i τ i = (5) 0 oherwise where: L : average number of ineracion over ime periods L : coun of muual ineracion of (, i ) in ime i S : sandard deviaion of ineracion beween (, i ) N : number of ime periods used o calculae Q : difference beween L and 1 L, i.e., L L 1 Regression analysis is a ool for he invesigaion of relaionships beween variables, and is maor use is predicion or forecasing [4]. Usually, he invesigaor seeks o ascerain he causal effec of one variable upon anoher. To explore he friendship beween friends, we assemble daa on he underlying variables of ineres (in our work, closeness, beweenness and evaporaion) and employ regression o esimae he quaniaive effec of hese hree variables upon friendship. In our work, we use he following regression model o esimae he friendship beween decision maker i and friend in ime : F = β + β C ( p ) + β C ( p ) + βτ + ε (6) i 0 1 c i 2 b i 3 i i L where β, β, β and β are parameers, ε 0 1 2 3 i is error erm, and E( ε i ) = 0, Cov( ε, ε ) = 0. i ik Var( εi ) 2 = σ, Afer he regression model is build, we can use his o measure decision maker s friend and selec required decision group. In our sysem, we selec op-n friends by ranking heir friendship and form he decision group. 3.2. Decision crieria discovering Our sudy used FDM for he screening of alernae facors [12]. The fuzziness of common undersanding of expers could be solved by using he fuzzy heory, and he efficiency and qualiy of quesionnaires could be improved. In our work, we followed ypical FDM process o implemen our sysem, bu made furher improvemen. To implemen FDM, we have o collec opinions of decision group firs. However, radiional quesionnaire survey for crieria collecing is ime consuming, so we design an online crieria collecing module o do he ob. To mainain basic requiremen of Delphi mehod, during he process, individual opinions are unknown o ohers. Afer collecing all he opinions, we calculaed he value of riangular fuzzy number of all facors and discovered he signi cance riangular fuzzy number of facors. By using simple cener of graviy mehod o defuzzify, he fuzzy weigh of each opinion can be convered o de nie value. Finally proper opinions can be screened ou as decision crieria from numerous facors by predefined hreshold value. 3.3. Decision crieria synhesizing In deermining evaluaion crieria phase, our sysem has screened he imporan facors conforming o a decision problem hrough FDM invesigaing expers crieria o se up he hierarchy archiecure. Here we modify ypical FAHP o calculae he weighs of individual crieria of a decision problem. Hsu and Chen [11] proposed a fuzzy similariy aggregaion mehod (SAM), in which similariies beween expers were collaed and fuzzy numbers assigned direcly o each exper o deermine he agreemen degree beween hem. Taking he degree of imporance of each exper ino consideraion, we modified he original weighing mehod as below. In [11], he average agreemen degree of exper E is given by n 1 A( E ) = Sk n (7) 1 k = 1 k 1780
where S k is he agreemen degree, and n is he number of expers. Besides, RAD is he relaive agreemen degree of exper E, which is formulaed as: AE ( ) RAD = (8) n A( E ) = 1 The relaive imporance of expers is formulaed as: r w =, = 1,2,..., n (9) n r = 1 Meanwhile, he consensus degree coefficien of exper E, = 1,2,... n is defined as: CDC = β w + (1 β) RAD (10) where 0 β 1. In our work, we improve he calculaion of relaive imporance of expers ( w ) and consensus degree coefficien of exper ( CDC ) o capure he spiri of social nework. For w, in he original definiion he weigh of he mos imporan exper is 1, ha is, r = 1. Then he kh exper is compared wih he mos imporan exper, and a relaive weigh r k is assigned. Since he decision group was seleced based on friendship F, in our i design he exper wih highes friendship index is considered o be he mos imporan exper wih r = 1, for all oher expers, rk = F F. Therefore, we can ik i reformulae he relaive imporance of expers as Fi w =, = 1,2,..., n (11) n F = 1 i 4. Experimen To furher prove he feasibiliy of our design, an empirical sudy alone wih sysem developmen was conduced. The procedures of our experimen are described as follows: 1. Build up regression model o compue friendship index beween decision maker and friends. 2. Issue decision problem by a user, hereinafer referred as decision maker, wihin a social nework. 3. Choose expers o form he decision group based on friendship index. 4. Build and exrac decision crieria online. 5. Synhesize decision crieria and form he alernaives. In he following secions, he imporan procedures are described in deail o clearly explain our sudy. To implemen our sysem, one of he mos popular social nework sies Facebook was seleced o be our experimen plaform. Firs we need o build up linear regression model o measure friendship index beween friends. For he purpose of collecing basic daa required by building regression model, a group of social nework users were invied o be our paricipans. Snowball sampling is a feasible way when sudying social nework issues [1], so i was used o consruc our experimen. A snowball sampling procedure wih S sages K names is defined as follows. A random sample of individuals is drawn from a given populaion. Each one in he sample is asked o name K differen persons, where K is a predefined number. For example, each person is asked o name K bes friends. The persons who were no in he random sample bu were named by individuals in i form he firs sage. Each of he individuals in he firs sage is hen asked o name K differen persons. This procedure repeas S imes o complee he sampling process. In he iniial sage, we drew 18 Facebook users randomly and divided hem ino hree groups by heir lifesyle. Afer filering ou hose users who are no willing o oin our experimen, we have suden group, office worker and random member groups. By using 3 (S) sages 3 (K) names snowball sampling we have 120 paricipans for each social nework (group). Tha is, a specific nework is formed by a coninuous node expending process unil a predefined maximum disance of connecions (i.e. 3 hops in our experimen) is reached. Afer filering he people no ineresed in our experimen, finally we have oal 181 unique paricipans, who form our experimenal social neworks. Of all he 181 paricipans, we randomly seleced 51 users o collec daa needed for consrucing our friendship regression model. The average year of usage is 1.6 years and he average number of friends is 155. An online survey package, including a cover leer explaining he research obecives and he quesionnaire was disribued o all he paricipans o survey heir online friendship. Since he closeness and beweenness cenraliy can be calculaed wihou paricipans help, our online quesionnaire lised all he friends of each paricipan, and asked hem o rae heir friendship. A 5-poin scale of nodding acquainance, ordinary friend, good friend, grea friend, bes friend is used o measure heir relaionship. Friends who are no raed by our paricipans are no included in our daase. Afer collecing he friendship daase, we used SPSS o build our regression model by enering friendship, closeness and beweenness cenraliy as base daa. This 1781
model was hen used o predic he friendship index a he ime when a decision problem is issued. Afer he regression model was buil, 8 random seleced users were invied o be our decision makers. In our experimen, hey can issue a decision problem and evaluae he effeciveness of decision crieria and alernaives. 73 problems abou consumer producs purchasing were issued during our wo monhs experimen, and hey were delivered o he decision group seleced by our sysem. Users wih op-5 friendship index were seleced as our decision group, and he processes repeaed when a problem was issued. Tha is, here are differen decision groups for differen problems. When a decision problem was presened o decision group, members can express heir opinions online. They were asked o choose some keywords from a predefined daabase WordNe [30], and describe why hese keywords relaed o decision problem. Follow he mehod proposed in our research, decision crieria and alernaives are generaed and presened o decision makers. Noe ha o precisely evaluae he effeciveness of our sysem, conen-based and collaboraive filering are used as benchmark mehods. The alernaives generaed from our work and benchmark mehods are presened o decision makers in he same page, and he links are lised randomly o minimize inerfere of presenaion order. To avoid informaion overloading, he firs wo alernaives of each mehod were seleced, and oal six alernaives were presened o decision makers for each problem. Finally, he click coun and say ime of each page linked o alernaives are recorded. If here exiss any overlap alernaives beween hree mehods, all he clicksream daa are duplicaed. 5. Discussion 5.1. Resuls and performance evaluaion The resul of social nework analysis is shown in Table 1, and he average friendship index in our experimen is 0.64. In our experimen we colleced clicksream daa of every alernaive presened, as shown in Table 2. As we can see, here are no significan differences beween number of clicks, bu he average say ime varies. To furher examine if here are significan differences beween say ime, a saisical mehod is required. Thus we used Tukey es for simulaneous muliple comparisons of our approach and all oher wo approaches, he es resul is presened in Table 3. As we can see, here is no significan difference beween he average say ime of collaboraive filering and conen based mehod, bu our approach has higher say ime. This means he decision makers spen more ime on our proposed alernaives han ohers. Since here is a srong endency for users o spend a greaer lengh of ime reading aricles of ineres o ha user [21, 32], we may conclude ha our approach is more effecive when compared wih oher mehods. Table 1. Saisics of beweenness closeness Beweenness Closeness Beweenness nbeweenness incloseness oucloseness Mean 89.928 0.279 66.727 66.728 Sd Dev 10.05 0.031 1.697 1.706 Sum 16277 50.518 12077.663 12077.729 Variance 100.998 0.001 2.879 2.909 Minimum 67.118 0.208 62.069 62.5 Maximu 124.462 0.386 71.146 72.289 Table 2. Saisics of clicksream daa Number of alernaives Collaboraive Filering 146 112.9775.41076.03399.02 1.92 Conen Based 146 124 1.0039.41227.03412.15 1.92 Our Approach 146 125 1.5215.27053.02239 1.01 1.99 Toal 438 361 1.1677.44670.02134.02 1.99 Table 3. Muliple comparisons of approaches 5.2. Decision group Number of click Mean say ime Sd. Deviaion Sd. Error Minimum Maximum 95% (I) Mehod (J) Mehod Mean Difference (I-J) Sig. Lower Bound Upper Bound Conen Based -.02643.815 -.1284.0756 Collaboraive Filering Our Approach -.54404 *.000 -.6460 -.4421 Conen Based Our Approach Collaboraive Filering Collaboraive Filering.02643.54404 *.815.000 -.0756.4421.1284.6460 Our Approach Conen Based -.51761 *.51761 *.000.000 -.6196.4156 -.4156.6196 *. The mean difference is significan a he 0.05 level. To es he influence of friendship index in he selecion of decision group, all he 73 decision problems were issued a differen ime period. Since he selecion rule of decision group was based on friendship index which is influenced parially by ime, we waned o know if he members change across differen ime. For example, when decision maker A issued his firs problem, a decision group consising of 5 members was buil, say ( M1, M2, M3, M4, M 5). A he ime when second problem was issued, anoher decision group, say ( M1, M2, M3, M7, M 8), was buil. In hese wo decision process, here were 10 users seleced as decision group, bu only 7 unique users since ( M1, M2, M3) was overlapped. The deail numbers are lised in Table 4. Furher analysis found ha decision makers wih a large number of unique users are more acive han hose who wih small number. However, no evidence showed ha here was significan difference of ime spen on he alernaives suggesed by differen decision groups even if here 1782
exiss differen number of unique users. This observaion showed ha our mechanism was sable enough and won be influenced by he uniqueness of decision group. Table 4. Users included in decision group Decision Maker Toal number of users seleced as decision group 6. Conclusion and fuure works In our paper, we inroduced ime facor ino social nework analysis. By using regression a friendship index calculaion model is proposed and served as our ool o predic friendship beween wo users in specific ime period. By equipping FDM wih online decision crieria mechanism, ime consuming problem of References Number of unique users A 40 31 B 45* 33 C 40 29 D 40 27 E 40 27 F 40 30 G 40 23 H 40 25 *In our experimen, B issued exra one quesion and we decided o include i in our experimen [1]. Y.Y. Ahn, S. Han, H. Kwak, S. Moon, and H. Jeong, Analysis of Topological Characerisics of Huge Online Social Neworking Services, In Proceedings of he 16h Inernaional Conference on World Wide Web, 2007, Albera, Canada, ACM, USA. [2]. B. Brehmer, Social Judgmen Theory and he Analysis of Inerpersonal Conflic, Psychological Bullein, 1976, 83(6), pp. 985-1003. [3]. L. Chi, C.W. Holsapple and C. Srinivasan, Compeiive Dynamics in Elecronic Neworks: A Model and he Case of Inerorganizaional Sysems, Inernaional Journal of Elecronic Commerce, 2007, 11(3), pp. 7-49. [4]. J. B. Copas, "Regression, predicion and shrinkage", Journal of he Royal Saisical Sociey, 1983, 45(3), pp. 311-354. [5]. L.C. Freeman, A Se of Measures of Cenraliy Based on Beweenness, Sociomery, 1977, 40(1), pp. 35-41. [6]. L.C. Freeman, Cenraliy in Social Neworks Concepual Clarificaion, Social neworks, 1979, 1(3), pp. 215-239. [7]. G. Groh and C. Ehmig, Recommendaions in Tase Relaed Domains: Collaboraive Filering Vs. Social Filering, In Proceedings of he 2007 inernaional ACM Conference on Supporing Group Work, 2007, Florida, USA, ACM, USA. convenional Delphi mehod was solved. Furhermore, an adopive SAM was also suggesed o furher improve he applicaion of FAHP on social nework relaed research. An empirical sudy furher proved he feasibiliy and effeciveness of our work. Being one of he pilo sudies in he developmen of social nework based decision suppor mechanism, alhough he research has reached is aims, here were some unavoidable limiaions. Firs, because of he ime limi, his research was conduced only on a small size of populaion who were he users of Facebook. Therefore, o generalize he resuls over differen social nework plaforms, he sudy should have involved more paricipans from oher social neworks. Second, o build regression model we needed basic daa. Alhough beweenness and closeness can be acquired direcly by analyzing exising connecions beween users, he iniial friendship daa requires manual collecion. To make our sysem more feasible, an auomaic friendship daa collecing mechanism would be necessary. Third, here are various measuremen used in he research of social nework analysis. In our work, only beweenness and closeness were considered. I would make our proposed friendship index more comprehensive by inroducing facors such as ie srengh and rus. [8]. J. Harmon and J. Rohrbaugh, Social Judgmen Analysis and Small Group Decision Making: Cogniive Feedback Effecs on Individual and Collecive Performance, Organizaional Behavior and Human Decision Processes, 1990, 46(1), pp. 34-54. [9]. S.R. Hilz and M. Turoff, The Nework Naion: Human Communicaion Via Compuer, The MIT Press, Cambridge, Massachuses, USA, 1993. [10]. V.B. Hinsz, Cogniive and Consensus Processes in Group Recogniion Memory Performance, Journal of Personaliy and Social Psychology, 1990, 59(4), pp. 705. [11]. H.M. Hsu and C.T. Chen, Aggregaion of Fuzzy Opinions under Group Decision Making, Fuzzy Ses and Sysems, 1996, 79(3), pp. 279-285. [12]. Y.L. Hsu, C.H. Lee and V.B. Kreng, The Applicaion of Fuzzy Delphi Mehod and Fuzzy Ahp in Lubrican Regeneraive Technology Selecion, Exper Sysems wih Applicaions, 2010, 37(1), pp. 419-425. [13]. A. Ishikawa, M. Amagasa, T. Shiga, G. Tomizawa, R. Tasua, and H. Mieno, The Max-Min Delphi Mehod and Fuzzy Delphi Mehod Via Fuzzy Inegraion, Fuzzy Ses and Sysems, 1993, 55(3), pp. 241-253. [14]. M.F. Kaplan and C.E. Miller, Group Decision Making and Normaive Versus Informaional Influence: Effecs of Type of Issue and Assigned Decision Rule, Journal of Personaliy and Social Psychology, 1987, 53(2), pp. 306. 1783
[15]. G.L. Klir and T.A. Folger, Fuzzy Ses, Uncerainy, and Informaion, Prenice Hall, Upper Saddle River, New Jersey, USA, 1988. [16]. D. Krackhard and L.W. Porer, When Friends Leave: A Srucural Analysis of he Relaionship beween Turnover and Sayers' Aiudes, Adminisraive Science Quarerly, 1985, 30(2), pp. 242-261. [17]. F.M. Krienen, P.C. Tu and R.L. Buckner, Clan Menaliy: Evidence Tha he Medial Prefronal Corex Responds o Close Ohers, The Journal of Neuroscience, 2010, 30(41), pp. 13906-13915. [18]. P.J.M. van Laarhoven and W. Pedrycz, A Fuzzy Exension of Saay's Prioriy Theory, Fuzzy Ses and Sysems, 1983, 11(3), pp. 199-227. [19]. P.R. Laughlin and A.L. Ellis, Demonsrabiliy and Social Combinaion Processes on Mahemaical Inellecive Tasks, Journal of Experimenal Social Psychology, 1986, 22(3), pp. 177-189. [20]. H.A. Linsone and M. Turoff, The Delphi Mehod: Techniques and Applicaions, Addison-Wesley, Boson, Massachuses, USA,2002. [21]. M. Moria and Y. Shinoda. Informaion Filering Based on User Behavior Analysis and Bes Mach Tex Rerieval, In Proceedings of he 17h Annual Inernaional ACM SIGIR Conference on Research and Developmen in Informaion Rerieval, 1994, Dublin, Ireland, Springer, USA. [22]. T.J. Murray, L.L. Pipino and J.P.V. Gigch, A Pilo Sudy of Fuzzy Se Modificaion of Delphi, Human Sysems Managemen, 1985, 5(1), pp. 76-80. [23]. E. Oe and R. Rousseau, Social Nework Analysis: A Powerful Sraegy, Also for he Informaion Sciences, Journal of Informaion Science, 2002, 28(6), pp. 441-453. [24]. J. Preece, Sociabiliy and Usabiliy in Online Communiies: Deermining and Measuring Success, Behaviour & Informaion Technology, 2001, 20(5), pp. 347-356. [25]. M.A. Rodriguez, Social Decision Making wih Muli-Relaional Neworks and Grammar-Based Paricle Swarms, In Proceedings of he 40h Annual Hawaii Inernaional Conference on Sysem Sciences, 2007, Hawaii, USA, IEEE, USA. [26]. G. Sasser, A Primer of Social Decision Scheme Theory: Models of Group Influence, Compeiive Model-Tesing, and Prospecive Modeling, Organizaional Behavior and Human Decision Processes, 1999, 80(1), pp. 3-20. [27]. C.L. Sreeer and D.F. Gillespie, Social Nework Analysis, Journal of Social Service Research, 1993, 16(1), pp. 201-222. [28]. H.T. Reis, W.A. Collins and E. Berscheid, The Relaionship Conex of Human Behavior and Developmen, Psychological Bullein, 2000, 126(6), pp. 844-872. [29]. M. Turoff, S.R. Hilz, H.K. Cho, Z. Li, and Y. Wang, Social Decision Suppor Sysems (SDSS), In Proceedings of he 35h Annual Hawaii Inernaional Conference on Sysem Sciences, 2002, Hawaii, USA, IEEE, USA. [30]. Princeon Uinversiy, WordNe: A Large Lexical Daabase of English, Available from: hp://wordne.princeon.edu/. [31]. O.S. Vaidya and S. Kumar, Analyic Hierarchy Process: An Overview of Applicaions, European Journal of operaional research, 2006, 169(1), pp. 1-29. [32]. G. Velayahan and S. Yamada, Behavior-Based Web Page Evaluaion, In Proceedings of he Inernaional Conference on Web Inelligence and Inelligen Agen Technology, 2006, Hong Kong, China, IEEE, USA. [33]. M.I. Yousuf, Using Expers' Opinions hrough Delphi Technique, Pracical Assessmen, Research & Evaluaion, 2007, 12(4), pp. 1-8. 1784