Towards a Reliable Framework of Uncertainty-Based Group Decision Support System

T owards a Re liable Fram ework of Uncertainty-based G r oup Decision Support Sy stem Junyi CHAI and James N.K . LIU Depa rtment o f Co mputi ng, The Hong Kong Poly technic University Hung Hom, Kowloon, Hong Kong, SAR { csjchai, csnkliu }@ comp.polyu.edu.hk Abstract-- This study proposes a f ramew ork of Uncer taint y-ba sed Gro up Deci sion S uppor t Sy stem (UGDS S). It provides a pla tfor m for multi ple c riteri a decision an alysis in six aspects incl uding (1) decisi on environm ent, (2) decision p roblem, (3) decision gr oup, (4) decision conflict, (5) decision sche mes and (6) group negotia tion. Based on multiple a rtificial in telligent technologies, this fra mework provides reliable support for the c omprehe nsiv e mani pul atio n of a pplic ati ons an d advanc ed de cisio n ap proaches throug h the de sign of an integra ted multi-a gents ar chitectur e. Keywor ds: Know ledge M anage ment; Un certainty ; MCDM; GDSS I. INTRODUCTION No wadays, si nce c ompanie s ar e usuall y worki ng in a rapidly changing and uncertain bus iness env ironment , more timely and accurate information are required f or decision -making, in order to im prove cu stomer satisf action, support profit able bus iness analys is, and increase their competitive advantages. In additio n to the use of data and mathematical models, so me m anagerial decisions are qualitative in nature and need j udgmental knowledge th at resides in hum an experts. Thus, i t is necessary to in corporate such knowledg e in developing Decision Support S yst em (DSS). A system that in tegrates knowledge f rom experts is called a Kn owledge-bas ed Decision Support Sys tem (KBDSS ) or an Int ellig ent Decision Support Sys tem (IDSS ) [1]. Moreover , tw o kinds of situations signi ficantly increase the complexity of decision proble m: (1) multiple participants involved in decision process; (2) decis ion-m aking under un certainty enviro nme nt. In this paper , we propose a f ramework of Uncertain ty-based G roup Decis ion Su pport Sy stem (UGDSS). Unlike existi ng DSS desi gns, this fra mework is based on multiagent technolog y and standalone knowle dge mana gement pro cess. T hro ugh the a dop tion o f agent technologies, this design provides an integ rated system platform to resolve the uncertainty problem in support of group Mult i-Criteria Decisi on Making (MCDM). Firs tly , we provide a general model of grou p MCDM. Then, we carry out an analysis on uncertainty- based group MCDM, and presen t our designi ng basi s of UG DSS. Thirdly , we propos e th e architectures and structures of UGDSS, including other two kin ds of knowl edge-rel ated sy stem compon ents: (1) Decision Resource MIS and (2) Kn owledge Base Manage ment Syste m (KB MS). The rest of t his paper is or g anized as follow s. Section 2 provides t he gen eral problem model of Group MCDM. Section 3 g ives th e uncertain ty analysis of grou p decision environm ent. Section 4 pr esents t he framework of UGDSS, i nclud ing t wo kno wledge -relat ed s ystem compon ents. Section 5 shows our conclu sion and out lines for future work . II. GROUP MUL TIPLE CRITERIA DECISION ANA L Y SIS A. Mu ltiple Criteria Decis ion Making Multiple Criteria Decision Making (MCDM) was derived from the Pareto Op timization Concept lo ng time ago. Aft er 50 years, Koopm ans [2] introdu ced the Eff icie nt Point in decision area. A t the same time, Kuhn and T ucker [3] introduced th e concept of V ector Optimi zation. Then, Charnes and C ooper [4] studi ed the model and applicati on of Linear Prog ramming in decis ion science. In 1972, t he International C onference on MCDM held by Coch ra ne and Zeleny [5] rem arked that the normative MCDM th eory had been developed as the mainstream o f decision sc ience. More recently , many applicable MCDM approaches h ave been used to design Decision Support System for solving specific domain problem s. The MCDM with certain information and under certain environment is called Classic M CDM. Major metho ds o f Classic MCDM can be roughl y divid ed i nto three categories: (1) Multiple Criteria Utility Theory , (2) Outranking Relations, (3) Preferen ce Disaggregation. 1) Multiple Crite ria Utility Theo ry Fishbu rn [6] and Huber [7] provided v ery s peci fic literature survey on Multiple Criteria Utility Theory . Besides, Keen e y an d Raif fa [8] publis hed a monograph which deeply influences the future development. 2) Outranking Relations The outranki ng relations approach aims to compare every couple of alternatives and then get s overall priority ranks, which mainly includes the E LECTRE method and the PROMETHEE m et hod. EL E CTRE was f irstly proposed by Roy [9] in 1960s . And, PROMETHEE method w as initially established by Brans [10]. Xu [1 1] extended PROMETHEE wi th a Superiority an d Inferiority Ranking (SIR) method which integ rated with the out ranking approach. 3) Pr efer ence Disaggr egation. 2010 IEEE International Conference on Data Mining Workshops 978-0-7695-4257-7/10 $26.00 © 2010 IEEE DOI 10.1109/ICDMW.2010.80 851 2010 IEEE International Conference on Data Mining Workshops 978-0-7695-4257-7/10 $26.00 © 2010 IEEE DOI 10.1109/ICDMW.2010.80 851 Jacquet-Lag reze et al. [12] prov ided a UT A method to maximize the approx i mation of the pref erence of decisi on makers by defining a set of ad ditive utility functions. Zopoun idis and Doumpos [1 3][14] developed the UT ADIS method as a variant of U T A for sorting problems, and exte nded the f ra mework of UT ADIS f or involving multi-participants cases called the MHDIS method. B. General Pr oblem Model of Gr oup MCDM The group MCDM problem involves multiple participants assessi ng alternatives based on multiple criteria. In order to facilitate the establishment and developm ent of MCDM su pport system, w e carry out a n anal ysis o f group MCDM to fo rm a ge neral model . Fig. 1 s hows our propos ed gene ral problem model for group MCDM . It genera lly c onta ins t hree d ecisio n set s: alternative sets ( i Y ), decision m aker sets ( k e ) and criterion sets ( j G ). It inv olves two kinds of we ights: deci sion maker wei ght s ( k w ) and criteri on w eights ( j ω ). Decision makers prov ide their individual deci sion matrix () k ij d and the wei ghts ( j ω ) of every criterion. In this figure, individual decision information is represented w it h a decision- plane ( k P ) includ ing blac k point s ( () k ij d ) and gre y points ( () k j ω ). Therefore, the group agg r egation process can be s hown as a plane- proj ection from individual decision plane k P to group- integrated decision pl ane () P K . Fig ure 1. G eneral pr oblem m odel o f Group M CDM III. UNCER T AINTY MUL TIPLE CRITERIA DECISION ANAL Y SIS Although C lassic MCDM already has qu i te a complete theory by now , it still cannot solve most MCDM problems in real w or ld. One main reason is that the decision inform ation are usually not provided completely , clearly or precisely in reality . In most cases, people have to m ake decisions in uncertainty environm e nt. Therefore, m any researchers pay more attention on this research branch of Uncertainty MCDM. Uncertainty MCDM is non-class ic, and can be treated as the extens ion of Cl assic MCDM. W e can generally divide the un certainty problems into three categories: (1) Stochas t ic ty pe (2) Fuzzy ty pe (3) Rough type. A comparison of th ese uncertaint y ty pes is sho wn in T able I. Accordingly , Uncertainty MCDM also has three res earch directions : Stochastic MCDM, Fuzzy MCD M and Rough MCDM. Recently , many approach es have been developed to solve these problem s (e.g. [15]). This paper w ill adopt this classification method to establish the fra mework of UGDSS including its subs ystems, several intellige nt agents and oth er functional applets. T ABLE I. The Comparison in Uncert ainty T ypes of Decision Problems MCDM Uncerta inty Objects Resear ch V ariabl e Major A pproach es Stocha stic type Possib ility of deci sion results is uncertainty (not ce rtain inc idence) decisio n attribute s in decisio n proble m 1. Ut ility Theor y 2. Proba bilit y Aggregat ion 3. Stoc hastic Si mulation … n Y 2 Y 1 Y 2 G 1 G m G Decision - Make r k e We i g h t o f j G 11 () ew 22 () ew () ll ew 1 ω 2 ω m ω () k ij d () k j ω Group Aggregat ion Process (Plane- Project ion) Indivi dual Decision - Plane k P Gro up-integr ate d decisio n Plane () PK 852 852 Fuzzy type The membership of objec ts is uncert ainty (not clar ity) The valu e of deci sion attrib ute in decisio n probl em Fuzzy sets (I ntuitionist ic Fuzzy , L inguisti c Fuz zy , … ) Rough t ype The gran ularit y of objec ts is uncert ainty (not accur acy) Decision schemes Rough sets (V a riab le Preci sio n RSs, Domina nce Based RSs , … ) 1) Stochastic MCDM Bayes t heory is proposed for stoch astic process whi ch can improve the objectivity and veracity in stoch a stic decision making. Then, Bernoulli [16] introduced the concept of Utility a nd Expected Utility H ypothesis Model. von Neumann a nd Mor genstern [17] conclu ded the Expected Utility V alue Theory , proposed the axiomatic of Expected Utility Model, and mathematically proved the results of maximized Expected Utility for decision maker . W ald [18] established the b asis of statistical d ecision problem , and applied th e m in the selection of stochast ical decision schemes. Blackwell and Girshi ch [19] integrated the subjective prob ability with the utility theory into a clear process to solve decision problem s. Savage [20] extended the Expected Utility Model, and Ho ward [21] introduced the sy stematical analysis approach into decision theory and dev eloped them from th eory and application aspects. 2) Fuzzy MCD M In 1965, Za deh [22] propos ed the Fuzzy Se ts which adopted the m e mbersh ip functions to represent the degree of mem bership from elements to sets. A tanassov and Garg ov [23][24] extended Zadeh’ s Fuzzy Sets concept into the Intuitionistic Fuzzy Sets (IFSs), and then as in the following, they extended IFSs in to the Interval-V alued Intuitionistic Fuzzy Sets (IVIFSs), which are describe d by a membership deg ree and a non-m e mbership degree whose values are interv a ls rather than real numbers. Based on t hese pioneerin g works, th eories of IFSs and IVIFSs have recei ved much atten tion from research ers. Until recently , some basic theorems such as Calcu lation Operators and Fuzzy Meas ures, hav e j ust been founded for vari ous applications[25][2 6]. In Chai and Liu’ s earlier work [15], a novel Fuzzy MCDM approach i s proposed based on t he Intu itionistic Fuzzy Sets (IFSs) theory to solve the real problem in Suppl y Chai n Mana ge ment (S CM). It first ly ap plie s IFSs to define and represent the fuzzy natural language ter ms which are used to desc ribe the in dividual decisi on values and the weights for decision criteria and decision makers. And then s i x main steps of this approach are presented to solve un certainty group MCDM problem . This work enriches the m et hod base of solving fuzzy MCDM problem , and can be im plemented in th e proposed UGDSS. 3) Rough MC DM Pawlak [27][28] system atically introdu ced the Rough sets theory . Then, Slowinski [29 ] concluded the past achievements of Rough s et s in theory and applications . Since 1992, th e annual International Conf erence on Rough Sets ha s bee n pla ying a very i mport ant r ole in promot ing the dev elop ment of Rough sets in t heory extension and v arious applications. More recently , Greco [30] propos ed a Dominance based R ough Sets theory which produ ces the decision rules with st ronger applicability . By now , Rough Sets t heory has been applied in decision analysis, process control, knowledg e discovery , machine learning, pattern recognition, etc. In UGDSS, the rough MCDM approach es are m ainly implemen ted by various s ubsystems with specific function modules, key intelligent a gents, and other deploy able applets. IV . UGDSS FRAMEWORK In this secti o n, we propose the f ra mew ork of UGDSS. Here, the term “Kno wledge” is a com prehensive concept, which in c ludes data , model, hum an knowledge and othe r forms of information, so lo ng as it can be used in uncerta inty group deci sio n maki ng. A. Uncertainty G r oup Decision Pr ocess and System Structur e In un certainty group decision proces s, we mainly consider three factors wh ic h increase the complexity of decision-m aking in reality: (1) Uncertain decis ion environm e nt, (2) Unstru ctured decision problem , (3) Complex deci sion grou p, and another issu e: Group unifi cation of decision conflict. This process provi des a mechanism to address th e t hree kinds of complexities and group conflict, which consist of six analysis stages: z Decision Environm ent Analysis z Decision Problem Analysis z Decision Group Analysis z Decision Scheme Analysis z Decision Conflict Analysis z Group Coordi nation and Decision Analysis Fro m decisio n-ma kers’ vie w , these sta ges ha ve t he basic logical sequence. Suppose t here is a MCDM probl em w ith com plex i nternal stru cture in volv ing multiple participants. W e fi rstly need to analyze the existing internal an d external environments, and f i gure out what are the decision conditions; whether the decision information is complete, cer tain and quantizable; what kinds of uncertainty type it belongs to. Secondly , the specific decision problems need to be analy zed, including ontological investi gation, problem representation and decomposition, etc. Thirdly , an ontological gro up analysis is required to reduce the complexity of human organiza tiona l str uctur e. Four thly , pe opl e need to establish problem-solvi ng solutions which may b e der ived fr om vari ous r eso urces i nclud in g previ ous problem -solving schemes in knowledge bas es, decision scheme s from do main ex p erts, or results of group discussion , etc. Fifthly , we need to integrate those 853 853 dispersive, multipurpo se, individual or incomplete decision opinion s into one or a set of applicable fin al decisio n results. Besides, the six sta ges mentio ned ab ove can moment arily call Negotiat ion Support Syst e m in conflict analysis stage for possi ble decision conflicts. From the view o f system process, each stage consists of s everal subsy stems w ith dif ferent fun ctions. For exampl e, we adopt th e ontologi cal problem analysis tools to represent, s crutinize and decom pose the complex decision problem . T hese su bs ystems are integ r ated in the UGDSS platform wi t h supports of interface technologies and intelligent a gent technologies. In t his design, parallel computation in subsys tems and middlew are is quite importan t, which ca n produce bett er system ef ficiency . Figure 2. UGDSS A rc hitec ture 1) Decision Envir onment Analysis Decision environment is an important factor wh ic h signif icantly influences oth e r decision stages. It may contain dif ferent aspects such as decision tar gets, decision principles, possi ble limitations, available resources , etc. More importantly , people need to analyze whether there are a ny unc ertain ty inf ormation . In th is pa per , w e defin e that the uncertainty decision infor mation consists of the following situations: z In form ation def icien cy z Information incompletion z Dyna mic in for matio n z Unclarity information z Inaccuracy information z Multiple uncertainties Although s everal MCDM approaches hav e been developed, it is not enough for solving co mplex uncertainty decisi on problem in reality . Therefore, one of our future w orks aims to establish an Uncertainty Enviro nme nt Anal ysis Sub -syste m (UE AS) to hand le uncertainty information, and t hen extend its capabilit y to solve oth er uncertainty MCDM problem s. 2) Decision Pr oblem Analysis W e can generally di vide decision proble ms i nto th ree categories: (1) St ructure, (2) Sem i-structure, (3) Non-structure. T o the first one, problem s are well org anized and represented for ontological an al ysis and decomposition. T o another two, problems are usually represented i n the form of text or interviewing dialogues. Therefore, these problem s need to be ontolog icall y represented an d described at th is stage. Some us e ful analysis techn ologies can be adopted including ontological analysis in Onto-bro ker [31], natural languag e process, etc. 3) Decision Gr oup Analysis Many decision problems in reality (such as great strate gic de cisio n of go ver nment o r ind ustr y , t he org anizational decision of lar ge corporation, etc.), involve multiple participants with complex human relationship or organiza tiona l str uctur e. A goo d gro up anal ysis can r esult in much efficien t decision process an d i mpartial decision resul ts. Grou p Support Sy stem (GSS) is used f or grou p analysis including decomposition, reorganization, character analysis, integration, etc. Some methods such as UGD SS Structur e Prog ramming Language supp ort DDS EIS ERP CR M SCM EC BPM … FF/ WF Securit y Su pp ort UIMS Multim edia Support Distribut ed Computing Wirel ess Support Visual Recognitio n Uncertai nty Analy sis Ontology Analy sis Auditor y Recognitio n Sensor y System Case based Reason ing NLP Computing GA Computing NN Computing … Networ k Protocol support Markup Language supp ort Applic ation Layer Intelligent Agents Layer T echnology Layer Probl em Analy sis Envir onment Analy sis Group Analy sis Scheme s Analy sis Conflict Analy sis Group Coor dination and Decision An alysis Uncert ainty GD SS Negotiation Support S ystem Group Support S ystem Onto-b roker UEAS UGD SS Archi tectur e KBMS DBMS MBMS KW Infer ence Engine Knowledg e-based Decision Re source MIS Stochas tic GDSS Rough GDSS Classi c GDSS Fuzz y GDSS DES Individual DSS 854 854 Double Selection Model [32] are a feasible approach to realize group analysis in GSS. 4) Decision Scheme Analysis Decision schemes are th e problem-solving solutions to specific decisi on problem . T hese sch e mes may be derived f ro m previ ous decision schemes reor ganized in Scheme Base; n e w problem-sol ving schem es establi shed by domain e xperts; solutions produced in group discus sion a nd negotia tio n; al l kind s of i nfor matio n on W eb or somewhere, et c. This stage is su pported by Domain E xpert Sy stem a nd correspon ding Decision Resource MIS. 5) Decision Conflict Analysis Decision conflict an al ysis is the core process in UGDSS. The conflicts may be derived at each stage of decision-m aking process. Theref ore, the subsy ste m at each stage may call the programs of Negotiation Support System (NSS) for conflict analysis. Chai and Liu’ s earlier work [32] prov ided a Group Arg umentation Model in order to solve complex decisio n conflicts. This model can be used to design and dev elop Negotiation Support Sy stem . 6) Gr oup Coor dination an d Decision Analysis This stage takes the responsibility for the integration of grou p opinion. Many methods can be used to solve th is problem lik e V ector Space Clustering , Entropy W eight Clustering, Intuitionistic Fuzzy W ei ght A verage (IFW A) method [26], W eigh ted Group Projection m ethod [15], etc. Besides, In dividual Decision Support Sy stem is a h elper of decisi on maker to develop t heir own opin ion, correspondin g with Domain E xpert Sy stems to form hi gh quality i ndividu al decision schemes. B. UGDSS Ar chitectur e Unlike existing des igns of DSS which mainly focus on specific problem do mains, the UGDSS architecture provides an in te grated system platform for complete decision an alyses and comprehensive application s. T he system architecture is sho wn in Fig. 2, which consists o f three layers: 1. The Application Layer 2. The Intelligent Agents Layer 3. The T echno log y La yer 1) Application L a yer a) Basic Function M odules z User Interface Manag ement System (UIMS) UIMS, as a su bsystem of UGDSS, is com posed of several programs and functional interface components in intelligent agent la yer such as natural language process, un certai nty analysis proces s, visua l reo rganizat ion f unct ion, etc. z Multim edia support Multimedia technologies are comprehensively used in UGDSS. The interfaces in application layer are related to many intelligent agents including V isual recognition, Audio recognition, etc. z W ireless support Many m obile application devices such as PDA, mobile phone, wireless facilities are used to support group deci sion-ma king z Security su pport In order to gu arantee the security of system and dat a transm ission , security support is indi spensible in system establish ment. Some main technolo gies like internal control mechanism, firewall, ID authentication, encryption techniques, digital signature, etc. b) Application D o main M odules The application domain m o dules aim to solve specific problem s in dif ferent domains . For example, Chai a nd Liu’ s Fuzzy MCDM method [15] is used to solve the problem of supply c hain partner s election. This application requires general d omain knowledge of Supply Chain Management. These application modu les as middleware of UGDSS can prov ide the necess ar y supports to various specific applicati o n dom a ins. Seve ral domains are given in following. z Financial/W eather Forecasting (FF/W F) z Director Decis ion Support (DDS ) z Enterprise Inf or mation System (EIS) z Enterprise Res o urce Planning (ERP) z Custom er Relati onship Mana ge ment (CRM) z Suppl y Chai n Mana geme nt (S CM) z E-Commerce (EC) z Busi ness Pr ocess M anage men t (B PM) 2) Intelligent Agent Layer z Sensor y system Sensor y systems, such a s vision systems, tactile system , and signal-process ing systems , provide a tool to interpret and analyze the collected kn owledge and to respond an d adapt to changes when facing diff erent en vironmen t. z Genet ic Al gorit hm (G A) co mp uting a gent Genetic Algorithms are s ets of computational pro ced ures, which lea rnt by p rod ucing o ffspri ng that are better and better as measured by a fitness functi on. Algo rith ms of t his t ype ha ve be en use d in decision-m a king process such as W eb search, financial forecasting, vehicle rou ting, etc. z Neura l Net work ( NN) c omp uting a ge nt A Neural Netw ork is a set of mathemati cal models whic h simula te the way a hu man br ain fu nctio ns. A typical intelligent a gent based o n NN technology can be used in st ock forecasting for decision making . z Un certain ty an alysis agen t This agent is used to an alyze the environm e nt and conditions of decision problem. z Case Based R easo ning (CBR) agen t Case Ba sed Rea soni ng is a means for solvi ng ne w problem s by us ing or adapt ing solutions of old problem s. It provi d es a foundation for reasoning, remembering, and learning. Besides, it simulates natural language expres sio ns, and prov ides access to organiza tio nal me mory . z Natural Languag e Process (NPL) computing 855 855 Natural L anguage Proces s (NLP) technology provides peop le the ability to communicate with a comp uter i n their nati ve la nguage . The goal o f NLP is to capture the meaning of sentences , which involves finding a representation for the sentences that can be connected to m ore general knowledge for decision making. Besides, all of these intelligent a gents with various group deci sion fu nctions may cons ist of dif ferent k i nds of knowl ed ge/inf ormation bases whi ch are unite d and embodied in Decision Resource Management Informati on System (DRMIS). This desig n can improv e the efficiency of inf o rmati on processing and the robustness of system. 3) T echnology Layer This layer provides the necessary sy stem supports to oth er tw o layers and DR MIS. It ma inly in clud es (1 ) programm i ng langu age support (VS .Net, C#, Java, etc.) (2) netw ork protocol support (HTTP , HTTPS, A TP , etc.) (3) marku p language support (HTML, XML, WML , etc). Besides, t he technol ogy layer also prov ides vari ous techn ology supports for cons tr uctin g the four Bas es, Inference eng ine, etc. C. Know ledge-r elated System Designs 1) Know ledge-based Decision Resour ce MIS fr amework Fig ure 3. K nowledge -based Decisio n Resource MI S Fr amewor k. Fig. 3 s hows the f ra mew or k of Kn o wledg e -based Decisi on Reso urc e MIS. It mainly c onsist s of four kind s of subsystem: KBMS, DBMS, MBMS, KW . In this syste m, di fferent ki nds o f infor matio n, kno wled ge, models and data interact together an d provide the supports to the whole UGDSS. z Data Base Mana gement System (DBMS) Generally , DS S needs a standalone database. Especially , this DSS is requ ired to solve the complex uncertain ty group decis io n problem s. Therefore, DBMS is a necessary component in UGDSS, which consists of a DSS databas e and a Data Mining System . A database is created, accessed, an d updated by a DBMS. A nd Data Mini ng Syst em is u sed to disc over kno wled ge fr om da ta resources. Many technologi e s are applicable to mining data, such as statistical approaches (Bay es’ s theorem, cluster analysis, etc), cas e-based reason ing, neural comp uting, ge netic a lgor ithms, etc. The se tec hnolo gies are developed as intelligent agents located in t he second layer in Fig. 2. z Model Base M anagement Sys tem (MBMS) MBMS mainly i ncludes Model Bas e and Model Anal ysis Syst em. M odel base co ntai ns routi nes a nd special statistics, financial forecasting, management science, and other qua ntitative models which provide the resources f or Model Analys is System. T urban [1] divided the m odels into four major ca tegories: Strategic, T actical, Operational, and Analytical. In addition, there are m o del building blocks and routines. Bas ed on th ese model resources, Model Analysis Sy stem is used to build blo cks; genera te t he new r out ines a nd r epo rts; up date and c hange model ; and m a nipulat e model data, etc. z Knowledge Bas e Management System ( KBMS) There are three ki nds of know ledge which w i ll be used in decis io n-making: (1) structure (2) semi-stru ct ure (3) non-s tr ucture. The structu ral knowledge is usually reorg anized in avail able models an d stored in model base. Much se mi- struct ural a nd no n-struc tur al kno wled ge ar e so complex th a t they cannot be easily repres ented and reorg anized. Therefore, more profess ional knowledge processing system called KBMS is required to enhance the capability of knowledge management. In ne xt section, we pre sent a de tailed kno wle dge ma nage ment pro cess i n … … T arget Base Probl em Base Principl e Base Condition Base Sch eme B ase Refine ment System Knowle dge Base Explanati on System Infer ence Engine Model Data Base Knowle dge Ware h ou se Model Base Man agement System Data Base Man agement System Knowle dge Base Man agement System Data Mini ng System Knowl edge based Deci sion Resour ce Management I nforma tion Sys tem Model Base Model A nalysis System Data Infor mation Knowle dge INTER- ACTION 856 856 KBMS. z Knowle dge W are house ( KW) In UGDSS, KW mainly contains multiple bases for classified s torage of decision knowledg e i ncluding decision problems, targets, principles, co nditions, schemes , etc. It is responsible f o r storage, ext raction, mainten a nce, interact ion and oth er knowledge manipulations. 2) Know ledge Management Pr ocess in K BMS Figure 4. Knowledg e Managem ent Process i n KBM S In this proces s, there are fi ve basic class es of knowledge manip ulation activities inc luding: acquisitio n, selection, generation, assi milation, and emissio n [33]. These activities are the basis for problem founding and solving, as well as being inv o lved at each stage in the decision making proces s. In this paper , w e provide a knowle dge mana gement p roc ess in KB MS (s ee Fig. 4). z Knowledge Res ource: S o me pos sible knowledge s ources include domain experts, books , documents , co mputer f iles, research reports, dat abase, sensors, an d any information avail able on the W eb. z Knowle dge Acquisiti on: This activ ity is the accumulation, tran s mission, and trans formation of document ed knowl ed ge resources or pro blem-solving scheme of e xperts. z Knowledge Repr esentation ： The acquired knowle dge is organi zed i n thi s acti vit y , whic h invo lve s preparation of a knowledge map and encoding th e knowledge in the kn o wledge base. z Knowledge Sel ection ： In kno wled ge re fine ment and explanation system, the kn o wledge is validated and verified until its quality is acceptable. There are three activities to refine a nd explain the acquired knowledge: (1) selection, (2) generation, (3) assimilatio n. In selection activity , syste ms select know ledge fro m inf or mation resources and m a king it suitable for subsequen t use. z Knowle dge Generation ： In this activity , knowledge i s produced based on the deci sio n in c ident by either dis covery or derivation f ro m exis ting knowledge. z Knowledge Assimilation ： In assimilation activity , this kno wledge refinement and explanation system alter the state of the decision makers’ knowledge reso urces b y distri buti ng and st oring t he acq uired , selec ted, or gene rate d kno wled ge [33 ]. z Infer ence Engine ： In knowl edge base, knowledge h as been org a nized properly and represented in a machine-unders tandable format. The inference engine can the n use the kno wled ge to in fer new conclusions from existing fact s and rules . T here are m a ny diff erent ways of representing hum an knowledge, incl uding Pr oduc tion rule s, Sema ntic net works, Logic statement, and Uncertainty inform at ion representation, etc. Here, know ledge is recogn ized and restored in knowledge base which al so co nducts the co mmunic atio n wit h other DRMISs. z Knowle dge Emiss ion ： This activity embeds knowledge into the outputs of KBMS, and input the useful knowledg e of specific decisional epis ode into UGDSS for fur ther d ecisio nal kno wled ge mani pula tio n activities including knowledge leadership, control and me asurem ent . Knowle dge Base Knowle dge Reso urces (Experts, book s, documents, Any i nfor m ati on on Web …) Knowle dge Repr esentati on Knowledg e Refinement S ystem Knowledg e Explanation S ystem Assimi lat ion Sele ction Genera tion Infer ence Engine z Case based rea soning z Rules based r easoning z Infer ence with Uncert ainty z … Knowledg e Manipulation In UGDSS z Knowle dge leade rship z Knowle dge Control z Knowle dge Measur ement Knowle dge Acquisition Knowle dge Emiss ion Other decision Resou rce M IS DBMS MBMS KW Stor age Int eraction Reorg anization, Re-st orage, et c. 857 857 V . CONCLUSION This paper prop oses the framework of Uncertainty Group Decision Support System (UGDSS) and oth er t wo kinds of knowledge-relate d system components: D eci sion Resourc e MIS a nd Kno wled ge B ase M anage ment S yste m (KBMS). W e first ly provi d e a gen eral problem model of group MC DM. And then, we carry out an analysis on uncertain t y group MCDM problem, an d present the bas i s of sy stem design on handl i ng th e uncertai nty decision -making. F i nally , we propose a set of detail ed designs of system architectures and structures for supporti ng a complet e uncertainty group decis io n-making process inclu d ing (1) env ironment analysis, (2) problem analysis, (3) g roup analysis, (4) schem e analysis, (5) group coordin ation analysi s and (6) decis ion conflict analysis. In future, w e will make e ff o rt on two directions. In system aspect, we need to develop multiple intelli gent agents, middlew are or subsy st ems which are integrated in UGDSS. In decision theory as pect, w e will consider h o w to develop more applicable uncertainty group decision -making approach es based on Fuzzy sets, R ough sets, Grey system theory and other un certaint y theories. ACKNOWLEDGMENT The author s woul d li ke to a ckno wled ge t he par tial supports from the GRF 5237/08E of th e Hong Kong Polytechnic University . REFEREN CES [1] E. T u rban, J. E. A ronson, T . P . Liang, Decision S u pport Syste ms and I ntellige nt System s, 7th ed., Pr entice Hal l, 2005. [2] T . Koopma s, “Analysi s of Produ ction as an E ffect Combinat ion of Activ ities,” Ne w Y ork: Jo hn W iley and So ns, 1951. [3] H. K uhn, A. T ucker , “No nlinear progr amming,” U niversity of Cali fornia Press, p p. 481 -492, 19 51. [4] A. Char nes, W . Co oper , “Manage ment Mode ls and I ndustrial Appl ications of L inear P rog ramming,” Ne w Y ork: John W iley and So ns, 1961. [5] J. L . Cochrane, M. Ze leny , “Multipl e Criteria D ecision Making,” Univers ity of Sout h Carolina Press, 1973 . [6] P . C. Fishbu rn, “Lexic ograph ic orders , utilit ies an d decis ion rules : A surv ey ,” Ma nagement Scienc e, vol. 20, pp. 1442-14 71, 1974. [7] G. P . Hu ber, “Method s for qua ntifyin g subjec tive p robabili ties and m ulti- attribute utilitie s,” Deci sion Scie nce, vol . 5, pp. 430-45, 1974. [8] R. L . Kee ney , H. Raiff a, Decisio ns W ith Mult iple O bjective: Prefer ences an d V alu e Tradeoffs, New Y ork: Wiley , 1 976 . [9] B. R oy , “C lass ement et c hoix en pres ence de p oint d e vu e multipl es: Le me thode ele ctre,” Revue Francai se d’Inf ormati que et de R echerch e Opera tionnell e, vol. 8 , pp. 57-75 , 1968. [10] J. P . Brans, P h. V incke, B. Mare schal, “How to se lect and how to rank proj ects : The PROMETHEE meth od,” Eu ropean Jour nal of Operat ional R esearch, vol. 24, pp. 228-238, 1986 . [11] X. Z . Xu, “ The SIR me thod: A superi ority and inferior ity r anking metho d for multipl e criteria decision m aking,” E urope an Journal of Operat ional R esearch , vol. 131, pp. 587 -602, 2001. [12] E. Ja cquet-Lagr eze, Y . Sisk os, “Asses sin g a set of a dditive u tilit y funct ions for mu lticrit eria dec ision-ma king: th e UT A Meth ods,” Europ ean Jou rnal of opera tiona l resear ch, vol. 1 0, pp. 151 -164, 1982. [13] C. Zo pouni dis, M. D oumpos, “B usin ess failur e predic tion using UT ADIS mu lti criter ia ana lysis, ” Jour nal of the Opera tion al Researc h Soci ety , vol. 50 , pp. 1138-1 1 48, 199 9. [14] C. Zop ounidi s, M. Doump os, “Build in g additi ve utili ties for mult i-group hierarchi cal dis criminat ion: th e M.H. DIS method,” Optim izati on Method s and Software, vol. 14, pp. 219-24 0, 2000. [15] J. Y . Chai, an d J. N. K. Liu, “ A novel mult icrit eria grou p decision makin g approa ch with int uitioni stic fu zzy SIR meth od,” Pro cee d ing i n The 8th I nternational Sym p osium on I ntell igent Auto mation an d Control (ISI AC), Kobe, J apan, S epte mber , 2010. [16] D. Bern oulli, “Exposi tion of a new theor y on th e measu rement of risk ,” Econom etrica , vol. 22, pp. 23-36 , 1954. [17] J. von Neumann, O. Morgenst ern, Theor y of Games a nd Econ omic Beha vior , Princ eton Universi ty Press, 1 944. [18] A. W a ld, S tatisti cal Deci sion Funct ions, New Y ork : John W ile y & Sons, 19 50. [19] D. Bl ackwell, M . A. Girs hick, Th eory of Gam es and Stati stica l Decisi ons, New Y ork: W i ley , 1954. [20] L . J. Sav age, T he Foundatio ns of S tatistics, N ew Y ork: J ohn W il ey & Sons, 1954. [21] R. A. Howard , “Dec isi on analysi s: Ap plied deci sion theory , ” Proceed in g in Th e 4th Inter nat ional C onferen ce of Op erat ion al Resear ch, New Y ork: W iley-Inter science, 19 66. [22] L . A. Z adeh, “F uzzy Se ts,” Info rmation a n d Contro l, vol. 8, pp. 338- 353, 19 65. [23] K. A tanassov , “Intuitionistic fuzzy sets,” Fuzzy Sets and Sy stems, vol. 2 0, pp . 87-96 , 1986. [24] K. Atan assov, G. Gargov , “Int erval-va lued i ntuiti onisti c fuzzy sets, ” Fuzzy Sets a nd Systems , vol. 3 1, pp. 343-349 , 1989. [25] K . Atanas sov , “Ope rators ov er interv al-value d intuition istic fuzzy sets, ” Fuzzy Sets a nd Systems , vol. 7 4, pp. 237-244 , 1994. [26] Z. S. Xu, “Intu iti onisti c preferen ce re lati ons and thei r applic ation in group decision making, ” Informati on Scien ces, vol. 177, pp. 2363-23 79, 2007. [27] Z. Pawlak , “Rou gh set,” Internati onal Jou rnal of C omputer and Informat ion Scienc es, vol. 11, pp. 34 1-356, 1982. [28] Z. Paw lak, Rough Se ts: Theor etical A spects of Reaso ning about Data, D ordrecht: Kluwer Academic Publishe rs, 19 91. [29] R. Slowinsk i, Intelli gent Dec ision Sup port: Handb ook of Appl ications an d Adv ances of Rough S ets theory , Dor drecht: Kluw er Academic P ublisher , 1992. [30] S. Greco, B . Matara zzo, R. Slowin ski , “Rough set s theory for multi criter ia decis ion analy sis,” Eur opean Jour nal of Operational Researc h, vol. 129 , pp. 1~47, 2001. [31] The URL of Ontobroker i s http ://www .on top rise. de/d eutsch /sta rt/prod ukte/on tobrok er/ [32] J. Y . Chai, and J. N. K . Liu, “A n onto logy -driven f ramewo rk fo r supp orting c omplex dec ision process, ” Proceed ing in The Inte rnational S ymposium on S oft Co mputing fo r Industr y (IS SC I) , Kobe, Ja pan, September, 2010. [33] C. Hol sapple, an d K. Jones , “Explor ing Primar y Activ ities of the Know ledge Chai n,” Kno wle dge Proce ss Manage ment, vol. 1 1, pp. 155–174 , 20 04. 858 858