关联规则挖掘在电子商务中的研究与应用

上海交通大学硕士学位论文关联规则挖掘在电子商务中的研究与应用姓名：马刚申请学位级别：硕士专业：计算机应用指导教师：张冬茉20080101vB2CAprioriAprioriviARTviiSTUDYANDAPPLICATIONOFASSOCIATIONRULESMININGINE-COMMERCEABSTRACTWiththedevelopmentofinformationandinternettechonology,E-commerceasanefficientnewcommercialmodelbecamemoreandmorepopular.Nowpeoplecancompletebusinessbytightlyclickthemouse.ButE-commercebringspeoplenotonlyconveniencebutalsoinformationoverload.Thissituationmakesithardforconsumerstofindtheproductsandservicestheywanted,especiallyfortheconsumersofB2CE-commerce.Thetechniquesofdataminingcansolvethis“Dataexplosion”problemwell.Associationrulesmining,asanimportantbranchofdatamining,issimpleandeasytoexplainandunderstand.Itcandescribetherelationshipbetweendataefficiently.Miningassociationrulesfromlargedatabaseshasbecomeahotareaofresearchinrecentyears.Youcanfindtheinstrinsiclinkbetweenthecommodityandcommodityandalsothecommodityandcustomerthroughminingassociationrules.IthasaveryimportantguidingsignificanceforthepersonalizedrecommendationinE-commerce,Enterprisemarketpositioningandsoon.viiiThisdissertationprovidesadetaileddescriptionoftheassociationrulesinthebasictheoryandalgorithmsoftheclassicalalgorithm-Apriorialgorithm.BecauseinthefieldofE-commerce,theApriorialgorithmhassomeproblemssuchaslowerefficiencyandredundantrules,thisdissertationpresentsassociationrulesminingalgorithmbasedonroughset,asthecharacteristicsofroughsettheory,thealgorithmcandealwiththeaboveissuesbetter.Thisdissertationalsofocusesontheimportanttrendofelectroniccommerce:personalizedservice,andintroducedsomepopularrecommendationtechnologiessuchascontentbasedrecommendation,collaborativefilteringbasedrecommendationandassociationrulesbasedrecommendation.InthedissertationitanalyzestheInadequateofthesetraditionalrecommendationtechnologiesandpresentsanewrecommendationtechnology,whichbasesontheARTNetworkandassociationrules.Thisrecommendationtechnologynotonlycanrecommendthecommoditiesbytheassociateditemsbutalsocanrecommendbythecharacteristicoftheonlineuser.Atlastofthepaper,therecommendationmechanismisemployedtoimplementaprototypeE-commercerecommendationsystemtoprovethefeasibilityandapplicabilityofthepresentedmethod.KEYWORDSdatamining,associationrule,E-Commerce,roughset,neuralnetworkiiiiv11.12090[1][2][3][4]RakeshAgrawal1993Apriori[5]internet21.21.3[6][7]RymonApriori31.4(1)Apriori(2)(3)1.5Apriori(ART)4(Associationrule)Agrawal1993Apriori90%()2.12.1()},...,{21nIIII=),...2,1(nkIk=kk-T(Transaction)TITIDDDTX⊆XYX⇒IYIX⊂⊂,φ=∩YXXYX⇒Y2.2()DXXXX(Support))(Xsupport:%100)(×=DXXsupportDDYX∩XYYX⇒)()()(YXPYXsupportYXsupport∩=∩=⇒2.3([8]))(Xsupport(minsup)X(frequentitemset)X52.4()YX⇒DDXYYX⇒(Confidence))(YXconfidence⇒%100)()()(×∩=⇒XsupportYXsupportYXconfidence(minconf)minsupYXsupport≥⇒)(minconfYXconfidence≥⇒)(⇒20%80%20%80%2.5()YX⇒D(ExpectedConfidence)YY2.6([9])XY112.2(1)D(2)D2-16Dminsupminconf2-1Figure2-1Thebasicmodelofassociationrulesmining2-1Dminsupminconf2.3:(1)BooleanassociationrulequantitativeassociationruleAge(x,”3035”)income(x,”42k48k”)=buys(x,”high_resolution_TV”)x(2)single-dimensionalassociationrule:buys(x,“IBMlabtop”)=buys(x,”high_resolution_TV”)Age(x,”3035”)income(x,”42k48k”)=buys(x,”high_resolution_TV”)(3)Age(x,”3035”)=buys(x,”laptopcomputer”)Age(x,”3035”)=buys(x,”computer”)(4)7Pccc’cc’2.4(1)(2)100D60X75Y40XYD0.30.6)66.0)(,40.0)((=⇒=⇒⇒YXconfidenceYXsupportYX2.4Y0.750.66XYPiatetsky-ShapiroPS[16]YX⇒(Interest)(Y)support(X)supportY)(XsupportYXIPS×−∩=⇒)(:(1)(Y)support(X)supportY)(Xsupport×=∩0=I(2)Y)(XsupportI∩∝(3))1)1(YsupportI(XsupportI∝∝8Piatetsky-Shapiro(X)supportXP=)((X)supportXP=)(),()(YXPYXP=∩(1)]1,0[]0,1[))(1))((1)(()()()(),()(:∪−∈−−−=⇒−IYPXPYPXPYPXPYXPYXItcoefficienφ0I-10=IXY0I1(2)],1[]1,0[),(),(),(),()(∞∪∈=⇒IYXPYXPYXPYXPYXI:Oddsratio(3)],1[]1,5.0[))()()(,)()()(max()(:∞∪∈=⇒IYXPYPXPYXPYPXPYXIConviction(4)],1[]1,0[)()(),()(:∞∪∈=⇒IYPXPYXPYXIInterest(5)]1,)()([])()(,0[)()(),()(:YPXPYPXPIYPXPYXPYXICosine∪∈=⇒(6)]25.0,0[]0,25.0[)()(),()(:∪−∈−=⇒IYPXPYXPYXIPS(7)]1,0[]0,5.0[))()|(),()|(max()(∪−∈−−=⇒IXPYXPYPXYPYXIe:Addedvalu(8)]1,0[),()()(),()(:∈−+=⇒IYXPYPXPYXPYXIJaccard8[17]2.52.5.192.5.22.5.3:2.5.4102.5.5(1)(2);(3)(4)2.6————113.1AprioriRakeshAgrawalRamakrishnanSkrikantApriori[5](Frequent-patterngrowthFp-growth)[10][11]Apriori3.1.1AprioriApriorik-)1(+k-1-1L1L2L2L3Lk-kLk-kAprioriApriori[12]XminsupXYXYX∪XYX∪Apriori[1]Apriori(1)kL1−kLk-kC][jliilj11−−∞=kkkLLC1−kL2−k1−kLiljl])1[]1[(])2[]2[(...])2[]2[(])1[]1[(−−∧−=−∧∧=∧=klklklkllllljijijiji]1[]1[−−klkljiiljl]1[il]...2[il]1[−kli]1[−klj(2)Apriori)1(−k-k-kkCc∈,kck-1−kckc)1(−k-1211−−∉kkLckkLc∉k-kck-kCApriori1Apriori3.1.2D3-12Apriori3-1DTable3-1TransactiondatabaseDTID123456{a,b,d}7{a,b}{a,b,c,e}8{a,c,f}{a,c}9{a}{a,b,c}10{b,c}{a,c,d}11{c,g}{a,b,d,e}D3-1(a)21-1L1L3-1(c)a,b,c,d,e22-D2-(d)22-3-1(f)3-1(f){a,c,d}{c,d}12-{a,c,d}{a,c,e}{a,d,e}{b,c,d}{b,c,e}{b,d,e}D3-1(g){a,b,c,d},{a,c,d}{a,b,c,e}{a,b,d,e}φ=4C2L3L3-213TID123456{a,b,d}7{a,b}{a,b,c,e}8{a,c,f}{a,c}9{a}{a,b,c}10{b,c}{a,c,d}11{c,g}{a,b,d,e}D{a}{b}{c}{d}{e}{f}19{g}167321C{a}{b}{c}{d}{e}967321L{a,b}{a,c}{a,d}{a,e}{b,c}{b,d}{b,