ROUGH SETS APPROACH TO HUMAN RESOURCE DEVELOPMENT OF INFORMATION TECHNOLOGY CORPORATIONS

ROUGH SETS APPROACH TO HUMAN RESOURCE DEVELOPMENT OF INFORMATION TECHNOLOGY CORPORATIONS SHINYA IMAI 1), CHE-WEI LIN 2), JUNZO WATADA 3) and GWO-HSHIUNG TZENG 4) 1), 2), 3) Graduate school of nformaton, producton and system,waseda unversty, Japan 4) Department of busness and entrepreneural management, kanan unversty, Tawan 1) shnyama@rur.waseda.jp, 2) heerowe1120@hotmal.com, 3) watada@waseda.jp, 4) ghtzeng@mal.knu.edu.tw Abstract: It s essental for IT corporatons to mprove compettve advantage and ncrease organzatonal performance. Employees are a key factor for a company s success. It s crucal to fnd or create a brand-new model n dealng wth human resource and customer relatonshp management, as well as to recognze whch employees characterstcs are nfluental n buldng relatonshps wth customers. The objectve of the paper s to clarfy what knds of features and behavours of the employees can create a good relatonshp wth customers. In the paper, rough sets model s used to deal wth vagueness/ambguty and uncertanty n the analyss of human resource and human relatonshp management, and can change a qualtatve problem nto a quanttatve one. The model wll gve useful nformaton by natural language and can provde gudelnes to a decson maker. The rough sets approach dfferentates between the two groups, and n the end we suggest some polces to mprove the qualty of human resource management, customer relatonshp management and the development of them. The proper management of employees and customers wll ensure the success of a project and the good performance of a corporaton. Keywords - human resource development, rough set theory, customer relatonshp management 1. INTRODUCTION Many researches underlne provdng mportant methods for human resource management. Snce n 80s people were product-orented; most employees just needed sklls to product, whch s the dea of job-based human resource management. Then Harvard Unversty psychology professor McClelland n 1973 frst challenged the dea of "work as the center" evaluaton. He ponted out utlzng competence sets nstead of ntellgence as assessment crtera, whch emphaszed the evaluaton dea and technology of "work as the center" [Rchard, 1997], the concept of competencybased evaluaton has led a revolutonary change n a modern evaluaton system. Ths dea regards an organzaton as a polymer of competence sets and ts focus on competence sets expanson and ts growth, especally sute to hgh dynamc and people-orented features n the era of knowledge economy [Yan et al., 2006]. So today employees do not need to pursue only ther jobs but also have to keep good relatonshp wth customers. When a company has a good relatonshp wth customers, t can ncrease the revenue, market share of the company, quck response of the market opportunty, customer loyalty to the company, and collect nformaton easly to ensure the corporaton resources that are used n a sutable way. For ths goal, a manager wll arrange employees to the rght place n the rght tme for satsfy company s customers. But when there s a quarrel between employees and customers, ths becomes a trade-off problem that means the company cannot satsfy both employees and customers at the same tme. Manager often sacrfces employees rght and satsfacton. Nowadays t s not correct any more. The objectve of ths research s to fnd out a compromsng soluton that satsfes both of employees and customers, and fnds out what knds of features and behavours of employees who can buld good relatonshps wth customers. The results of ths research wll be used to gude organzatons as they are useful n provdng a good strategy for human resource management and customer relatonshp management IJSSST, Vol. 9, No. 2, May 2008 31

Ths paper s organzed as follows: Secton 2 revews research efforts. The mathematcal model employed here s brefly llustrated n Secton 3. Secton 4 s spent to explan the problem tracked n ths paper. Its results are dscussed n Secton 5. At the end several remarks of ths paper s gven n Secton 6. 2. RELATED RESEARCH EFFORTS 2.1. Human Resource Management Storey (1995) gave a good defnton of human resource management (HRM) as t s a dstnctve approach to employment management whch seeks to acheve compettve advantage through the strategc deployment of a hghly commtted and capable workforce usng an array of cultural, structural and personnel technques. In recent years, by connectng HRM wth strategc management and corporaton performance, researchers emphaszed on organzatonal performance under the postve lnkage wth HRM promotes to organzatonal performance [Truss, 2001][ Pfeffer and Vega, 1999]. If the companes have good HRM, then t wll have good organzaton performance, so how to nspre worker s motvaton and passon s the key pont of a busness to survve n an IT ndustry. The research [Pfeffer, 1994] showed that hgh motvaton and strong commtment of employees wll lead to hgh busness performance n the long run. HRM should be nfluenced to brng effectveness for an organzaton, whch can be used as a strategy for decson makers to attach work to ther staff. Casco (1992) suggested that organzatons nowadays must gan compettve advantage through the effectve utlzaton of ther HRs; the human power of an IT ndustry s complex, because employees strve wth balancng ther personal development and loyalty for a company. People change ther job frequently, because sometmes they want to receve hgher salares, benefts or statements, but ths s a very large loss of the company because the company spent tranng costs, the tme for employees, the busness nformaton and the relatonshp and contact data wth the costumers, all of those thngs wll loss when those experenced workers are hred by ther company s compettors. So HRM fnds out some worker s features whch can reduce ths problem. Human resource, n other words, well-nformed capable ctzenry can mprove the total ablty of an organzaton, a socety, a government agency and vrtually of a country, of a naton [Khan, 2003]. For example, plannng and needs based recrutment keep staff costs down; mert-based selecton procedures should mprove the qualty of staff [World Bank, 1997], companes can have qualty, effectve and low cost employees though HRM at the same tme. 2.2. Soft Computng The hardware and software of computer enable us to deal wth large data n a short tme. Huge amount of data have been collected and stored n a database, tradtonal ad hoc mxtures of statstcal technques and data management tools are no longer adequate for analyzng ths vast collecton of data [Sushmta et al., 2002]. Many busnesses need huge collectons of data lke: fnancal nvestment, human resource management, customer relatonshp management, producton and nventory management etc. But we are faced wth another problem, that s, how to analyze such a huge number of data. There are many researches who use data mnng and soft computng to fnd out meanngful nformaton from a large database [Sushmta et al., 2002], collect current data mnng practce nclude the followng. (1) Classfcaton: classfes a data tem nto one of several predefned categorcal classes. (2) Regresson: maps a data tem to a real valued predcton varable. (3) Clusterng: maps a data tem nto one of several clusters, where clusters are natural groupngs of data tems based on smlarty metrcs or probablty densty models. (4) Rule generaton: extracts classfcaton rules from the data. (5) Dscoverng assocaton rules: descrbes assocaton relatonshp among dfferent attrbutes. (6) Summarzaton: provdes a compact descrpton for a subset of data. (7) Dependency modelng: descrbes sgnfcant dependences among varables. (8) Sequence analyss: models sequental patterns, lke tme-seres analyss. The goal s to model the states of the process generatng the sequence or to extract and report devaton and trends over tme. 32

Data mnng s wdely used n many researches, and varous soft computng methodologes have been appled to handle dfferent challenges posed by the data mnng [Sushmta et al., 2002]. There are many methods such as fuzzy logc, neural network, genetc algorthms, genetc programmng and rough sets. Each of them can analyze a problem n ts doman, those methodologes can be used together to solve complex problems, and more and more researches combne those methods to fnd new crtcal features. That result s more adaptve for our real world as comparng wth tradtonal technques. Current researches fnd conventonal data mnng methods stll have weak ponts. Those methods focus on dscoveryng algorthm and vsualzng technques. But through data mnng t s easy to fnd out a huge number of patterns n a database, where most of these patterns are actually useless or unnterestng to the user [Sushmta et a.l, 2002]. Rough set theory suts to analyss of dfferent types of uncertan data and rough set can deal wth large data to reduct superfluous nformaton, we also can fnd extractng knowledge form the rules. al., 2002,], travel demand analyss [Goh and Law, 2003], data mnng [L and Wang, 2004], the research proposal of a general approach for a progressve constructon of a rule-based assgnment model to solve the lnear programs [Azb and Vanderpooten, 2002]. Based on rough sets theory, the research by Shyng et al. (2007) addressed the effect of attrbutes/features on the combnaton values of decsons that nsurance companes make customers needs satsfed [Shyng et al., 2007]. Rough set theory can unfy wth fuzzy theory [Lech P., 2003] and s transformed from the crsp one to a fuzzy one, called Alpha Rough Set Theory [Quafafou, 2000]. Another paper dscussed the preference-order of attrbute crtera needed to extend the orgnal rough sets theory, such as sortng, choce and rankng problem [Greco, S. et al., 2001] and extends prevous research by employng a development of rough sets theory, namely the varable precson rough sets (VPRS) model busness falure predcton [Beynon and Peel, 2001], The rough sets theory s useful method to analyze data and reduct nformaton n a smple way. 3.2. Method 2. ROUGH SETS 3.1. Researches Rough sets theory has many advantages. For nstance, t provdes effcent algorthms for fndng hdden patterns n data, fnds mnmal sets of data (data reducton), evaluates sgnfcance of data, and generates mnmal sets of decson rules from data. It s easy to understand and offer straghtforward nterpretaton of results [Pawlak, Z., 1996]. Those advantages can make the analyss easy, that s why the rough sets approach s appled wdely n many researches. The rough sets theory s of fundamental mportance n artfcal ntellgence (AI) and cogntve scence, especally n the areas of machne learnng, knowledge acquston, and decson analyss, knowledge dscovery nductve reasonng, and pattern recognton n databases, expert systems, decson support systems. Rough sets theory s developed by [Pawlak, 1982; Pawlak, 1984]. It has been appled to the analyss of many ssues, ncludng medcal dagnoss, engneerng relablty, expert systems, emprcal study of materal data [Jackson et al., 1996], machne dagnoss [Zha et In IT corporatons, human resource s an mportant ssue, and how to manage employees and customer relatonshp s full of ncompleton and uncertanty. Human resource management s about people, ncludng a manager, staff and customer. Much of human knowledge s expressed n natural language and a natural language s bascally a system for descrbng perceptons. Perceptons are ntrnscally mprecse, whch reflect the bounded ablty of sensory organs, and ultmately the bran, to resolve detal and store nformaton [Zadeh, 2005], those perceptons are hded n the human language. The tradtonal research methods are dffcult to measure the real meanng of the human percepton. Fuzzy set and rough set theores turned out to be partcularly adequate for the analyss of varous types of data, especally, when dealng wth nexact, uncertan or vague/ambguous knowledge [Walczak and Massart, 1999]. Both the fuzzy set and rough set theores deal wth the ndescrbable and percepton knowledge. The most dfference between them s rough set theory does not have membershp functon so that t can avod pre-assumpton and subjectve nformaton n analyss. 33

Rough set theory provdes a new dfferent mathematcal approach to analyze the uncertanty, and wth rough sets we can classfy mperfect data or nformaton easly. We can dscover the results n terms of decson rules. So n ths research, we use rough set theory to analyze the human resource problem. 3.2.1. Informaton system: Generally, an nformaton system, denoted by IS, and defned by IS = (U, A, V, f), where U conssts of fnte objects and s named a unverse and A s a fnte set of attrbutes {,,.,,}. Each attrbute a belongs to a set A, that s, a A. fa : U Va, where V a s a set of values of attrbutes. It s named a doman of attrbute a. Example 1. Attrbutes decson objects a1 a2 a3 D1 object1 3 1 3 1 object2 1 2 3 2 object3 3 1 3 1 object4 1 2 3 2 object5 2 2 4 2 object6 1 2 2 1 object7 2 2 4 1 object8 1 2 2 2 object9 1 2 3 2 object10 2 2 4 2 U= {object1, object2, object3, object4, object5, object6, object7, object8, object9, object10} A= {a1, a2, a3} a1= {1, 2, 3} a2= {1, 2} a3= {1, 2, 3, 4} D1= {1, 2} 3.2.2. Lower and Upper Approxmatons: A method to analyze rough sets s based on the two basc concepts that are lower and upper approxmatons of a set as shown n Fgure 1. These are two crsp sets. In Fgure 1 some squares are ncluded n the crcle, whle the others are not. The squares ncluded completely n the crcle are called a lower approxmaton, whereas the squares partally and completely ncluded n the crcle are called an upper approxmaton. Let X be a subset of elements n unverse U, that s, X U. Now we consder a subset P n V,.e., a P V a. The low approxmaton of P, denoted as PX, can be defned by the unon of all elementary sets x contaned n X as follows: { nd ( p ) [ ] } PX = x U x Χ where x s an elementary set contaned n X, = 1, 2,..., n. The upper approxmaton of P, denotes as PX, can be defned by a non-empty ntersecton of all elementary sets x contaned n X as follows: ={ [ ] ( ) } The boundary of X n U s defned n the followng: PNX = PX uuu PX Fgure 1 shows conceptually low approxmaton and upper approxmaton. Example 2 a1 a2 a3 {object1, object3} 3 1 3 {object2, object4, object9} 1 2 3 {object5, object7} 2 2 4 {object6, object8} 1 2 2 If we are nterested n subset X of fve objects: X= {object1, object2, object4, object5, object9} The elementary sets presented n example2, whch are also contaned n X, are: {object1, object3}, {object2, object4, object9} So the Lower approxmaton s PX= {object1, object3}, {object2, object4, object9} To calculate the upper approxmaton of subset X, one has to fnd all elementary sets n Example 2 whch have at least one element n common wthn the subset X, whch are: {object1, object3}, {object2, object4, object9}, {object5, object7} 34

Fgure 1. Upper and Lower Approxmatons of Set X. So the Upper approxmaton s PX= {object1, object3}, {object2, object4, object9}, {object5, object7} Lower approxmaton Upper approxmaton Boundary of X n U s PNX= {object1, object3}, {object2, object4, object9}, {object5, object7} - {object1, object3}, {object2, object4, object9} = {object5, object7} 3.2.3. Core and Reduct of Attrbutes: Core and reduct attrbute sets are two fundamental concepts of a rough set. Reduct can mnmze subset and make the object classfcaton satsfy the full set of attrbutes. The core concept s commonly used n all reducts [Shyng et al., 2007]. Reduct attrbutes can remove the superfluous attrbutes and gve the decson maker a smple and easy nformaton. There may be more than one reduct attrbutes. If the set of attrbutes s dependent, we are nterested n fndng all possble mnmal subsets of attrbutes whch have the same number of elementary sets [Walczak and Massart, 1999]. The reduct attrbute set affects the process of decsonmakng, and the core attrbute s the most mportant attrbute n decson-makng. If the set of attrbutes s ndspensable, the set s called the core [Walczak and Massart, 1999], whch s defned as θ φ, formulas θ and φ are called condton and decson, respectvely [Walczak and Massart, 1999]. Though the decson rules we can mnmze the set of attrbutes, reduct the superfluous attrbutes and group elements nto dfferent groups. In ths way we can have many decson rules, each rule has meanngful features. The stronger rule wll cover more objects and the strength of each decson rule can be calculated n order to decde the approprate rules. 1. 4. AN EMPIRICAL STUDY ON HUMAN RESOURCE DEVELOPMENT OF IT CORPORATIONS The questonnares has been bult based on the several analyss we provded n IT ndustry [Toyoura et al., 2004] and n analyss of accdents [Watada et al., 1998]. In ths research we obtaned answers of 47 questons of 167 employees from IT companes. These questons are named attrbutes whch characterze each employee. When employees have good customer relatonshp t always help a company earn more proft and realzes a good brand. Nevertheless, t s dffcult to fnd out whether employees wll have a good customer relatonshp. Ths ssue s mportant n every corporaton, but a manager s hard to recognze who has good relatonshp and nteracton wth consumers, Therefore, we place a focus on the answer of the queston: do you have a good human relatonshp wth your customers? Let us denote an answer of employee to ths queston ( = 1,2,...,167) as A. The objectve of ths research s to clarfy what knd of pattern of other answers to the questonnares results n some value of A, that s, yes, I have a good relatonshp wth customers or no, I have not a good relatonshp wth them. We nvestgate all the answers of 167 employees of IT companes and fnd out the latent structure of the answers as managers and companes can provde effectve functons to motvate ther staffs. RED( P) A COR( B) = RED( P). 3.2.4. Decson Rules: Decson rules can also be regarded as a set of decson (classfcaton) rules of the form: a d where k j a means that attrbute k a wth value I, k means the decson attrbutes and the symbol denotes propostonal mplcaton. In the decson rule d j In the analyss, we frstly processed the data that obtaned from questonnares by ROSE [Predk and Wlk, 1999], [Predk et al., 1998]. Table 1 shows the lower and upper approxmatons obtaned by a rough set analyss. Ths result has accuracy 1.000. Ths means the target set s defnable on the bass of an attrbute set [Pawlak et al., 1998]. Attrbute A ( = 1,2, L,167 ) do you have a good human relatonshp wth your customers? s named as a decson attrbute. The values that A ( = 1,2, L,167 ) 35

Class number 1 2 Table 1. Lower and upper approxmatons Number of objects 146 21 Lower approxm aton 146 21 Upper accuracy approxm aton 146 21 1.000 1.000 takes are 1 (yes) and 2 (no). There are 146 YESs and 21 NOs. Table 1 llustrates that the upper and lower approxmatons are equvalent. Therefore, there s no uncertanty n the classfcaton of classes D = 1 and D = 2. When decson rules are obtaned, the decson rules can help a decson maker to get more nformaton about human resource. 4.1. Decson Rules Decson tables n APPENDIX A show the coverng rate and the elements of rules. It helps us to know those features more clearly, and each rule has ts own elements. Those elements are the features of these rules. We want to fnd out the typcal rules that can cover most of employees and help decson makers to get the deal pcture of employees behavor, n ths way manager can dfferentate features nto many groups and each group has ts own polces. Through the ROSE n Fgure2 we can obtan the results, whch are shown n APPENDIX A, where we fnd 16 rules. There are 9 decson rules that have good relatonshp wth customers and 7 decson rules that have bad relatonshp wth customers. The total coverng rate s 98.8%. So ths decson table stands for employees behavors n the IT corporatons. APPENDIX A shows that rule1 covers 59.59% of the whole employees. More than half of people n these IT companes have such features when they have good relatonshp wth customers. Rule 10 covers 28.57% of the whole employees. Such behavors of those employees have ther customers who may not be satsfed. Rule 10 covers only 28.57% people because n the databases of 167 employees, only 21 employees do not have good nteracton wth consumers. The coverage of 28.57% s stll meanngful n our result. 5. RESULTS AND DISCUSSIONS In rules 1 and 4 we can understand when employees n the IT corporatons have good relatonshp wth customers. They also have postve thnkng and behavors. They understand a whole system or organzaton and contact wth people of ther customer s company, lke to jon the fnancal statements and thnk t s useful to attend a busness protocol tranng, know board members and nternal operatng rules of the company, and also have good relatonshps wth other members n the same company. Through those features we can know they are very actve n ther company. So managers should pay attenton to how to make them contrbute contnually and ncrease ther motvaton to make more beneft for ther company In rules 10 and 12 the employees wth bad relatonshp wth ther customers are those who do not understand the contract, the whole system and organzaton of ther customer company, jon n none of scheduled tranngs, lectures or semnars and carrer-up, never contact wth any board members of customer companes, have no knowledge of databases about customers, and have none of solutons n troubles of the collaboraton wth a maker. Those negatve features mean employees do not care about ther jobs and do not want to make achevements. The manager of such employees must fnd out the reason and push them back to work. Those postve and negatve decson rules let us know what knd of features and behavors can brdge between customers and the company. Maslow s herarchy of needs s often depcted as a pyramd consstng of fve levels: psychologcal needs, safety, love/belongng, esteem, and self-actualzaton. People have good relatonshp wth customers s n the level of love /belongng. So they wll want to seek for next level whch s more esteemed. People need to engage themselves to gan recognton and have actvtes that gve the person a sense of contrbuton. As people try to become better, they work hard to fnd ther abltes. Those people always want to learn new thngs and know more nformaton about ther company, because they have already realsed they are one of the team. Therefore, such employees wll play a central role n ther jobs. Employees, who have a bad relatonshp wth customers stll n the safety level, only care about themselves whether they wll be employed or not. It means they thnk everythng for themselves. Therefore, they do not want to work hard because t means nothng to ther safety. 36

Fgure 2: ROSE wth the result of decson rule. They do not thnk they are company members. Therefore, those people have no nterest n dong anythng out of ther duty. A manager has to assure securty of ther employment. It means to protect the rght of ther job and does not mean that the manager fres people easly. They should be placed nto the love/belongng level. In ths way, the manager can have effectve staff. The manager also has to take care of that people who have good relatonshp wth customers, help them do ther work correctly, sometmes gve them some challenges and rewards to ncrease ther motvaton. No matter whether the employees have good or bad relatonshp wth customers, they are the human resource of the corporaton. So Manager can gve those two groups wth dfferent functons. But t s always not easy to dvde people nto groups, so manager can recognze ther staffs accordng to the result of rough set theory. In APPENDIX A when the decson attrbute equal to one, t means people have good relatonshp wth customers and they have those feathers: they recognze and understand a whole system and organzaton of your customer s company, they understand the board members, they lke to attend any tranngs or semnars for acqurng the knowledge of fnancal statements, they have a good human relaton wth ther bosses, senor members, and colleagues wthn a company...etc, those features can gve a guld for manager to evaluate the employee. When the decson attrbute equal to two, the staff have those features: they do not understand the contract wth the company whch they work n, they don t attend scheduled tranngs, lectures or semnars for busness protocol and carrer-up after jonng ths company, they can t contact wth board members of ther customer s company, those negatve work atttude wll let the manager gve them another dfferent way of management. In ths research we gve a clear functon to make staffs dfferentable. 37

6. CONCLUDING REMARKS Through ths research we can understand human resource management and customer relatonshp management are mportant n every corporaton. The rough sets theory separates superfluous factors from mportant elements n our decson table and successfully deals wth the human resources and customer relatonshp management, besdes we fnd out latent logc of features and answers to manage employees and customers. Those useful nformaton can be adopted n IT corporatons, to help them to provde rght servce at rght tme whch satsfes ther customers wthout sacrfces employees rghts, n another word, ncreases the satsfacton of employees and customers at the same tme. REFERENCES Azb, R., and Vanderpooten, D. 2002, Constructon of rule-based assgnment models, European Journal of Operatonal Research, Vol. 138, No. 2. Pp274-293. Beynon, M.J., and Peel, M.J. 2001, Varable precson rough set theory and data dscretsaton: an applcaton to corporate falure predcton, Omega, Vol. 29, No. 6. Pp561-576. Casco, W.C. 1992, Managng Human Resources: Productvty, Qualty of Work Lfe, Profts, McGraw- Hll, New York. Goh, C., and Law, R. 2003, Incorporaton the rough sets theory, Chemometrcs and Intellgent Laboratory Systems, Vol. 47, No. 1. Pp1-16. Greco, S., Matarazzo, B., and Slownsk, R. 2001, Rough sets theory for multcrtera decson analyss, European Journal of Operatonal Research, Vol. 129, No. 1. Pp1-47. Jackson, A.G., Leclar, S.R., Ohmer, M.C., Zarko, W. and Al-kamhw, H. 1996, Rough sets appled to materals data, ACTAMater, Vol. 44, No. 11. Pp4475-4484. Khan, Md. S. 2003, Readng promoton: Perspectve Bangladesh, CDN LAO newsletter, No. 48, November, http://www.ndl.go.jp/en/publcaton/cdnlao/048/481.ht ml. Lech P. 2003, Rough methodology: A rough set paradgm for unfyng rough set theory and fuzzy set theory, Fundamenta Informatcae, Vol. 54, No. 1. Pp67 88. L, R., and Wang, Z.O. 2004, Employees behavors, European Journal of Operatonal Research, Vol. 157, No. 2. Pp439-448. Mtra, S., Pal, S.K., and Mtra, P. 2002, Data mnng n soft computng framework: a survey, IEEE transactons on Neural Networks, Vol. 13, No. 1. Pp3-14. Pawlak, Z. 1982, Rough sets, Internatonal Journal of Computer and Informaton Scence, Vol. 11, No. 5. Pp341 356. Pawlak, Z. 1984, Rough classfcaton, Internatonal Journal of Man Machne Studes, Vol. 20, No. 5 Pp469 483. Pawlak Z. 1991, Rough Sets, Kluwer Academc Publshers. Pawlak, Z. 1996, Rough set and data analyss, Proceedngs of the Asan11-14 Dec.. Pp1 6. Pawlak, Z.. 1999, Rough classfcaton, Int. J. Human-Computer Studes, Vol. 51, No. 15. Pp369-383. Pawlak, Z. 2004, Decson networks, In: Rough Sets and Current Trends n Computng by Shusaku Tsumoto, Roman Slownsk, Jan Komorosk, Jerzy W. Grzymala- Busse (Eds.), Lecture Notes n Artfcal Intellgence (LNAI), Vol. 3066, No. 1. Pp1 7. Pawlak, Z., 2005, Rough sets and flow graphs, Rough Sets, Fuzzy Sets, Data Mnng and Granular Computng, LNAI Vol. 3641. Pp1-11. Pawlak, Z. Wong, S.K.M. and Zarko, W. 1988, Rough sets: Probablstc versus determnstc approach, Internatonal Journal of Man-machne Studes, Vol. 29, No. 1. Pp81-95. Pfeffer, J. 1994, Compettve Advantages Go through People, Boston, Harvard Unversty Press. Pfeffer, J., Vega, J.F. 1999, Puttng people frst for organzatonal success, The Academy of Management Executve, Vol. 13, No. 2. Pp37 38. Predk, B., Slownsk, R., Stefanowsk, J., Susmaga, R. and Wlk, Sz. 1998, ROSE - software mplementaton of the rough set theory, In: L.Polkowsk, A.Skowron (eds.), Rough Sets and Current Trends n Computng, Lecture Notes n Artfcal Intellgence, Vol. 1424, Sprnger-Verlag, Berln. Pp605-608. Predk, B., and Wlk, Sz. 1999, Rough set based data exploraton usng ROSE system, In: Z.W. Ras, A. Skowron (eds.). Foundatons of Intellgent Systems, Lecture Notes n Artfcal Intellgence, Vol. 1609. Sprnger-Verlag, Berln. Pp172-180. Quafafou, M. 2000, α-rst: a generalzaton of rough set theory, Informaton Scences, Vol. 124, No. 4. Pp301-316. 38

Rchard J.M. 1997, Everythng you wanted to know about competency modelng, Tranng & Development, Vol. 51, No. 8. Pp73-77. Shyng J.Y., Tzeng G.H., and Wang F.K. 2007, Rough set theory n analyzng the attrbutes of combnaton values for nsurance market, Expert System wth Applcatons, Vol. 32, No. 1. Pp56-64. Truss, C. 2001, Complextes and controverses n lnkng HRM wth organzatonal outcomes, Journal of Management Studes, Vol. 38, No. 8. Ppl1122 1149. Walczak, B. and Massart, D.L. 1999, Rough set theory, Chemometrcs and Intellgent Laboratory, Vol. 47, No. 1. Pp1-16. World Bank. 1997, World Development Report: The Changng Role of The State, Oxford Unversty Press. Yan A.I., Lu Y., Lu Z.C, Chen Z. 2006, Study on Human Resource Nche Concept and Evaluaton Indexes, Management Scence and Engneerng, ICMSE. Pp1290 1295. Zadeh L.A, 2005 Toward a generalzed theory of uncertanty (GTU)- an outlne, Informaton Scences, Vol. 172, No. 1-2. Pp1-40. Zha, L.Y., Khoo, L.P., and Fok, S.C. 2002, Feature extracton usng rough set theory and generc algorthms an applcaton for the smplfcaton of product qualty evaluaton, Computers & Industral Engneerng, Vol. 43, No. 4. Pp 661-676. Toyoura,Y., Watada, J., Yabuuch, Y., Ikegame, H., Sato, S., Watanabe, K., Tohyama, M. 2004. Fuzzy Regresson Analyss of Software Bug Structure, Central European Journal of Operatons Research (CEOR), Vol. 12, No. 1. Pp. 13-23. Watada, J., Tanaka, T., Arredondo, A.R. 1998, Analyss of Safety from the perspectve of Macro- Ergonomcs, Journal of Japan Ergonomcs Assocaton. Vol. 34, No. 6, 1998. 39

APPENDIX A Rule member Rule(1) 59.59% Decson attrbute The mnmal coverng rule Value Coverng rate 1 Q29 Do you recognze and understand a whole system and organzaton of your customer s company? Q30 Do you understand the board members? Yes Q33 Q36 Do you lke to attend any tranng or semnar for acqurng the knowledge of fnancal statements Do you have a good human relaton wth your bosses, senor members, and colleagues wthn a company? Q40 Do you understand the nternal operatng rules of the company? Yes Yes 0.520 Yes Yes Rule(2) 45.21% 1 Q29 Do you recognze and understand a whole system and organzaton of your customer s company? Q39 Do you understand the contract wth the company that you work n? Yes Yes Q2 Is t useful n busness that you attended busness protocol tranng? Yes Yes 0.640 Rule(3) 23.97% 1 Q36 Do you have a good human relaton wth your bosses, senor members, and colleagues wthn a company? Q4 Dd you attend a scheduled tranng, lectures or semnars for busness protocol and carrer-up after jonng ths company? Yes 0.700 None Q46 To whom do you consult when you experence any troubles n The boss collaboraton? Q48 Are you male or female? Male Q52 How many books do you read n a month? About 5 books Rule(4) 10.27% 1 Q18 Regardng the method to learn sklls of software products, how about your general product knowledge level of software products? Q38 Q42 Are you satsfed of the promoton and award programs wthn the company? Regardng the contracted rules and collaboraton wth a computer maker, do you understand the yearly measures and polces of the maker? Low 0.724 Yes No Rule(10) 2 Q39 Do you understand the contract wth the company that you work n? No 0.880 28.57% Q4 Dd you attend a scheduled tranng, lectures or semnars for busness protocol and carrer-up after jonng ths company? Q31 Can you contact wth board members of your customer s company? No Q50 What s your job ttle? None General level Rule(11) 23.81% 2 Q29 Regardng the knowledge of the company system and organzaton, do you recognze and understand a whole system and organzaton of your customer s company? No Q33 Do you lke to attend any tranng or semnar for acqurng the knowledge of fnancal statements f you have any chances? No 0.910 Q9 Do you have my database about your customers? Out of my duty Q45 Dd you fnd any solutons n these troubles of the collaboraton wth a maker? No Yes 40

Rule(12) 19.05% 2 Q45 Dd you fnd any solutons n these troubles of the collaboraton wth a maker? Q5. How do you fnd busness manners of other employees around you? Bad Q28 Q43 Do you follow and backup your proposal wth specfcaton and estmaton after gvng t to a customer? I propose rather estmaton. Do you understand dealng rules (prce rate of products) wth a maker and the sales promoton program? Q53 Do you prefer to play a sport? Yes No 0.934 I propose rather estmato n. No Rule(13) 2 Q30 Do you understand the board members? No 0.976 14.29% Q33 Do you lke to attend any tranng or semnar for acqurng the knowledge of fnancal statements f you have any chances? Q43 Do you understand dealng rules (prce rate of products) wth a maker and the sales promoton program? Q12 How about your general product knowledge level of I seres? Low Q32 Regardng the knowledge of fnancal statements, dd you learn how to read and analyze the fnancal statements? Yes No Yes Total 0.988 41

SHINYA IMAI et. al: : ROUGH SET APPROACH TO HUMAN RESOURCE DEVELOPMENT Shnya Ima: Jonng IBM Japan Ltd. n Aprl. 1971. Leavng IBM Japan Ltd. n March. 2006. It takes charge on the small and medumszed scale customer and busness partner etc. n whle servng. He s PhD applcant on Management Engneerng, Graduate School of Informaton, Producton and Systems, Waseda Unversty snce Aprl 2006. Hs research nterests nclude the mprovement of a corporate qualty and the employee qualty, especally, the research of the relaton between QWL(Qualty of Workng Lfe) and the personnel tranng. Che-We Ln: receved a B.C. degree n nternatonal trade from Chnese Culture Unversty, Tawan n 2005, he receved two master degrees; one s n MBA from Kanan Unversty, another s n nformaton, producton and system from Waseda Unversty. Junzo WATADA: He receved hs B.S. and M.S. degrees n electrcal engneerng from Osaka Cty Unversty, Japan, and hs Ph.D degree from Osaka Prefecture Unversty, Japan. Currently, he s a professor of Management Engneerng, Knowledge Engneerng and Soft Computng at Graduate School of Informaton, Producton & Systems, Waseda Unversty, Japan. He was a recpent of Henr Coanda Gold Medal Award from Inventco n Romana n 2002 and a fellow of both SOFT and BMFSA. He s a contrbutng prncpal edtor, a co-edtor and an assocate edtor for varous nternatonal journals, ncludng JSCE of IMECH E and IJBSCHS. Hs professonal nterests nclude soft computng, trackng system, knowledge engneerng and management engneerng. Gwo-Hshung Tzeng: Gwo- Hshung Tzeng was born n 1943 n Tawan. In 1967, he receved the Bachelor s degree n busness management from the Tatung Insttute of Technology; n 1971, he receved the Master s degree n urban plannng from Chung Hsng Unversty; and n 1977, he receved the Ph.D. degree course n management scence from Osaka Unversty, Osaka, Japan. He was an Assocate Professor at Chao Tung Unversty, Tawan, from 1977 to 1981, a Research Assocate at Argonne Natonal Laboratory from July 1981 to January 1982, a Vstng Professor n the Department of Cvl Engneerng at the Unversty of Maryland, College Park, from August 1989 to August 1990, a Vstng Professor n the Department of Engneerng and Economc System, Energy Modelng Forum at Stanford Unversty, from August 1997 to August 1998, and a professor at Chao Tung Unversty from 1981 to the present. Hs current research nterests nclude statstcs, multvarate analyss, network, routng and schedulng, multple crtera decson makng, fuzzy theory, herarchcal structure analyss for applyng to technology management, energy, envronment, transportaton systems, transportaton nvestment, logstcs, locaton, urban plannng, toursm, technology management, electronc commerce, global supply chan, etc. He has got the Pnnacle of Achevement Award 2005 of the world and had got the natonal dstngushed char professor and award (hghest honor offered) of the Mnstry of Educaton Affars of Tawan and three tmes of dstngushed research award and two tmes of dstngushed research fellow (hghest honor offered) of Natonal Scence Councl of Tawan. Fellow IEEE Member (From September 30, 2002). He organzed varous nternatonal conferences. He s edtors-n-chef of Internatonal Journal of Operatons Research, Internatonal Journal of Informaton Systems for Logstcs and Management, and so on. 42