DOC

Evaluation Model for Scientific Quality Based on Rough Sets and Its Empirical Study

By Chad Peters,2014-02-18 20:12
9 views 0
Evaluation Model for Scientific Quality Based on Rough Sets and Its Empirical Study

    Evaluation Model for Scientific Quality

    Based on Rough Sets and Its Empirical

    Study

    Vo1.15No.3JournalofSouthwestJiaotongUniversity(EnglishEdition)Ju1.2007 ArticleID:1005-2429(2007)03-0247-05

    RoughSetsandItsEmpiricalStudy

    LIUDun(刘盾)h

    JIANGChaozhe(蒋朝哲

    HUPei(

    )LIULi

    )

    (刘丽)

    0n

    J.SchoolofEconomicsandManagement,SouthwestJiaotongUniversity,Chengdu610031,China

    2.SchoolofAdultEducation,SouthwestJiaotongUniversity,Chengdu610031,China Abstract

    Byanalyzingthequestionnairesrecollectedfrom74differentgovernmentdepartmentsinChengdu,China,anevaluationmodel

    forscientificqualityofcivilservantswasdevelopedwiththeroughsettheory.Intheempiricalstudy,aseriesofimportantrules

    weregiventohelptocheckandforecastthedegreeofthescientificqualityofcivilservantsbyusingthereductionalgorithm,andthe

    totalaccuracyofpredictionwas93.2%.

    KeywordsRoughset;Rulereduction;Scientificqu~ity

    Introduction

SincePawlakproposedtheroughsetstheory

(RST)in1982,thismethodologyhasbeendevel

    opedquicklyandusedwidelyinmanyfieldssuchas industry,iatrology,informationscience,etc.For instance,Aijuneta1.usedRSTtopredictwater

    supplyconsumption;Tsunoto.JusedRSTinclinical

    diagnosis;ChouchoulasandShen['usedRSTin

    informationfilteringandinformationretrieva1.Allof theirexploratoryworkareimportantandeffective. Furthermore,intheeconomicandmanagementfield, RSTwasappliedtobusinessfailureprediction, databasemarketing[andfinancialinvestment~.

    Received20o6_()9O7

    FoundationitemDoctoralFundofMinistryofEducation ofChina(No.20060613019)

    BiographyLDun(1983),candidate.Hisresearch

    ofinterestisinroughsetstheoryanditsapplication,andman- agementdecision-making

    }Correspondingauthor.Te1.:+86?13882268673;E?mail: newton83@163.com

    Thestudyreportedinthispaperisaresearch efforttoimplementthePlanofCivilScientificQuality (Theplanof2049)atthebeginningofthe21st century.Questionnairesweredesignedanddelivered tothecivilservantsof74governmentdepartmentsin Chengdu;thenaseriesofruleswereproducedbyrule reductionalgorithmtoconsmactanevaluationmodel forscientificquality.

    ExperimentalMethodandTechni-

    calRoute

    Toevaluatethescientificqualityofpeople,a questionnaireinvestigationwasadopted.Theques

    tionnaireisdividedintothreeparts:thefirstpart includessomebasicinforillationsuchassex.age, educationallevel,post,etc.oftheinformant;the secondpartincludestheacquiringchannelofscientific informationandtheinterestingdegreeofscientific knowledgeoftheinformant;thethirdpartistotest andevaluatethescientificqualityoftheinformant. Sevenkindsofknowledgewerechoseninourexpefi

    JournalofSouthwestJiaotongUniversity(EnglishEdition) ment,includingtheknowledgeinscience,astronomy andgeography,biomedicine,computer,economics, managementdecisionmaking,andhumanities. Totally,2212questionnairesweresenttothe civilservantsfrom74governmentdepartmentsinthis Getting

    newrules

    survey,and2192validquestionnaireswererecollec

    ted.ThetechnicalrouteoftheevaluationmodelWas designedasFig.1toanalyzetheknowledgestructure ofthesequestionnaires.

    Repeatingthe

    processiftherule

    setisnotperfect

    Fig.1Evaluationframewo~foradministrativemembers'scientific

    qualitybasedonroughsetmodel

    2EvaluationModel

    AninformatiOnsystemisdefinedasapairS=

(,A),whereUisanonemptyfinitesetofobjects,

    i.e.theuniverse(the74governmentdepartmentsale regardedastheuniverseinoursmdy);A=CuDisa nonemptyfinitesetofattributes.whereCdenotes thesetofconditionattributes,Ddenotesthesetof decisionattributes,andCnD=f2j(inourstudy,7 kindsofknowledgealechosenasCandthelevelsof scientificqualityischosenasD,respectively).In addition,akindofknowledgeinadepartmentcorres

    pondstOavaluef(,a),and,:U×AVisan

    informationfunctionsuchthat,(,a)?forall?

    Uanda?A,whereV=u.The4-tupleS=(,

    A,,representsaninformationsystem,whichcor

    respondstothedecisiontableofscientificquality basedonRST[?.

    For?U.thedecisionrulesaleexpressedas R(x):^((C,(,C));C?C)

    (d,,(,d);d?D),

    whereA((C,,(,C));C?C)and(d,f(,d);d?

    D)arethepremiseandconclusionoftherole, respectively.

    Let[]cdenoteapartitionbythecondition attributesC,whichsatisfies

    []c=(Y?;VC?C,,(,C)=,(Y,C)).

    Let

    U/C={,,,C,}.

    Forany

    Cj?U/C(f=1,2,,,1),

    thecardinalnumbersinCtaledenotedbyII=.,

whichmeansthattheequivalenceclassCjincludes.

    departments.

    Let

    Fd(C)=(,(,d);?Cf)=

    {Wil,Wi2,,W'},

    whichcorrespondstoJdecisionvaluesofJdepart

    mentsinCj.

    Fortheequivalenceclass,therealeJdecision

    LIUDuneta1./EvaluationModelforScientificQ?ZBasedonRoughSetsandItsEmpiricalStudy249 rulesasfollows:

    R(C):A((c',(c,Ci));c?C)

    (d,wft;d?D),(k=1,2,,_).

    Then,thereliabilityofR(C)isdefinedas cc=,

    whereDft=(?U;f(,)=wrt).Thus,therelia

    bilityoftheruleP(R(C))istheratioofthe numberoftheCiwiththedecisionattributevaluew": tothetotalvalueofCf.

    LetlcfldenotethecategorynumberofR(C), and

    lCfnDitl

    denotethecoverdegreeofR().IfCiDthe

    reliabilityofR(C)equals1.

    3EmpiricalStudy

    Byanalyzingtherecollectedquestionnaires,the scoresaboutdifferentkindsofknowledgeindifferent

    departmentscanbeobtained:

    n0

:l'2,'74;y-l,2,'7),

    ,l"

    whereruvdenotesthemarksofthevthknowledgein theuthdepartment;,ldenotesthenumberofcivil servantsintheuthdepartment;aupvdenotesthemarks ofthepthcivilservantoftheuthdepartmentinthe vthkindofknowledge.n.andareknown.

    Then,theaveragescoreratioisdefinedas w

    .

    ,("=l,2,,74;1,=l,2,,7),

    wheremydenotesthefullSCOreofthevthknowledge inthequestionnaire,whichisalsoknown. Theaveragescoreratiocanbeeasilyacquired throughcomputationbasedonthedatabase,which canhelptoconstructthedecisiontable.Beforecon

    structingthedecisiontable,wediscretizedtheoriginal datafirstaccordingtotheprincipleofpartition,as showninTablel_

    Denotetheknowledgeofscience,astronomyand geography,biomedicine,computer,economics, managementdecisionmakingandhumanitiesbycl,

    c2,c3,c4,c5,c6andc7,respectively.Then,welet =

    {ll,12,13,14,l5,16}

    correspondtocertainknowledgeofthedepartments. Table1Theprincipleofpartitionaccordingto experts'advice

    Classifythe74departmentslnto3kindsaccord

ingtoclusteranalysismethod.The3kindsareevalu

    atedbyexperts,whibhis

    ',d={1,2,3}

    andmeansthatthescientificqualityofthedepartment isgood,average,unsatisfactory,respectively.The detailedinformationisshowninTable2,where1,d denotesthedecisionvalueand1,d?',d.

    Duetotheredundancyanddependencyofthese attributes,thecommonalgorithmforattributereduc

    tionandvaluereductionisusedtoreducethedecision table.Then,13mainruleswhichhaveahighrelia? bilityandahighcoverdegreeareacquiredbyusing thesoftwareRSESdevelopedbyWarsawUniversity inPoland.asshowninTable3.

    Inordertovalidatethereliabilityandvalidityof the13mainrules,somenewtestsamplesshouldbe selected.Inthispaper,the74samples(trainingsets) arereusedastestsetsbyRSESandthedetailedresults areshowninFig.2.Althoughthetestresultseelns oversatisfactory,itcanstillsatisfythetestrequestof ruleseffectivelyE...

    FromtheresultsinFig.2,weknowthat,forthe 7departmentswhosedecisionvalueisequalto1,the accuracyofpredictionis100%;forthe57depart- mentswhosedecisionvalueisequalto2,theaccura- cyofpredictionis91.2%;fortheremaining10 departmentswhosedecisionvalueisequalto3,the accuracyofpredictionis100%.Allinal1.thetotal accuracyofpredictionis93.2%.whichsuggeststhat thepredictioniseffective.

JournalofSouthwestJiaotongUniversuy(EnglishEdition)

    Table2Thedecisiontableof74depar~ents No.ClC2C3C4C5C6C7dNo.ClC2C3C4C5C6C7d 114Isf3f2f2f4238f414f3fl1313f33 21414f3f2f2f213239f41412f2f2f2f32 313f3f2f2f2f2f334014f4f31212f32

    4f4f4f2f3f2f3341141412f2f2142

    514Is131212f2f4242f414f3f2f2f2142 614Isf3f312f214243f31412f2fl12f31 7f4fsf3f2f214244f3f4f31212f32

    8IsIsf2f4f212142451414f2f21212f32 9f4f4f3f2flf214246f414f3f2f2f2132 1Of4f4f3f2f2f2f3247f416.[2f2f2fl132 11f4f4f2f2f2f213248f4Is12f2f2f2132 12f4f412ll13ll13249l31412l312l2l32 1314121212f21325of31412f2flf2131 14f4f3f2flf2f3151141412f2flf2142 1514Isf3f2f214252f3f2f2f2f2f32

    16f3fsf3f2f2f3f3353f4f5f2f31212f22 17f4f4f3f2f2f213254141613131212132 18f4f3f2f2f2f3255Is1313f2岛岛Is3

    19岛岛f2f2fl1212156141412121212132 20l3l2l2f2flf213157f4f3f2flf2f2f32 21fsf3f2f213258f41412flf212f42 22fsf4f3f2f2f213259f314f2f2f2f2f32 23f4f4f3f2f2f21326of41413f2f2f2f32 24f4f4f2f2f2f213261f3f4f2f2fl12131 25f4f4f2f2f2f213262f4131313fl12132 26f4f4f2f2f2f313363f414f3f2f2.[2f32

    27f4f4f2f2flf213264f41412f21212f32 28f4f3f2f2f2f2f3265f4f412f2f3f2f43 29f4f4f3f2f2f213266f4f3f212f2f32

    3Of3f4f2f2f21326714f5f3f2.[2f32

    31f4f2f2f2f213268f4fsf3f2f2f32

    32f4f6f3f3flf21326914Isf3f31212143 33f3f2f2flf213170f314f2f21212132 34f4f6f2f2f212143711414f2f212f2f32 35fsfsf31412f214372f4Isf3f31212f32 36f4fsf3f2f2f2142731316f31312f2fs2 37f4f412/2f2f3274f'14f3f31212132 Table311Ie13mainrules

LIUDuneta1./EvaluationModelforScientificQual~BasedonRoughSetsandItsEmpirical

    Study251

    9隧掰.je瀚,..:0a

    5250570.9121

    Actual01001011

    007711

    ;Trueposjtirerate10.15671

    Tot.aZr,tLmbe]~Oftestedoh1et0:74

    TOta1.acCt,IlZ:ac~r0.932

    ToA1COV~Eaqe:1

    Fig.2Thetestresultsofmainrules 4Conclusion

    (1)Forthedepartmentwhosescientificquality isgood(vd=1),theknowledgeofeconomicsshould

    beperfect(c5=,I),theknowledgeaboutmanage

    mentanddecisionmakingshouldbeperfectorexcel

    lent(c6=Zlor12),andtheotherknowledgemustbe

betterthanaverage.

    (2)Forthedepartmentwhosescientificquality isaverage(Vd=2),theknowledgeabouteconomics andmanagementshouldbeperfectorexcellent(c5= Zlorf2,c6=Zlor12),andatmostoneoftheother knowledgeshouldbelessthanunsatisfactory(atmost oneattributeisequaltol5or16).

    (3)Forthedepartmentwhosescientificquality isunsatisfactory(=3),

    ?theknowledgeabouteconomicsormanage

    mentislessthangood(c5orc6islessthanZ3); ?theknowledgeabouteconomicsandmanage? mentareexcellent(c5=12orc6=Z2),butmorethan twokindsofknowledgeseemstobeunsatisfactoryor poor(atleasttwoattributesareequaltolsor16). Inaword,theknowledgeabouteconomicsand managementplaysaveryimportantroleinevaluating thescientificqualityofgovernmentdepartment, whichisinaccordancewithourexperience. Inaddition,somedecisionruleswhosereliability degreesarelessthan1arefoundinthissystem.The reasonisthattheattributevaluesofdifferentdecision attributesaredifferent,butthedifferenceisconcealed bythediscretizationofdata,whichresultsinthe inconsistentrules.Thetotalaccuracyofpredictionis upt093.2%byRST,whichshowsthattheproposed methodisreasonableandvalidfordgingthescien

    tificqualitiesofcivilservants.

    ?

    References

Report this document

For any questions or suggestions please email
cust-service@docsford.com