基于FP-樹的頻繁模式和長模式挖掘.pdf

1、鄭州大學(xué)碩士學(xué)位論文基于FP樹的頻繁模式和長模式挖掘姓名：王秉政申請學(xué)位級別：碩士專業(yè)：計算機軟件與理論指導(dǎo)教師：范明2003.5.1ABSTRACTDataminingresultsfromthesituationdescribedasdatarichbutinformationpoorAsanimportanttaskofdatamining，frequentpattemminingisemployedinmanyapplicati

2、onssuchasminingassociations，correlations，causality，sequentialpatterns，episodes，multidimensionalpatterns，maxpattems，partialperiodicity，emergingpatterns，etcForalongtime，acategoryofApriori—likealgorithmshasbeenadoptedformin

3、ingfrequentpatternsButtheysufferfromtakingmanySCanSofdatabasesforhugenumberofcandidatepaRemoccurrencefrequenciescheckingFPGrowthalgorithmadoptspatternfragmentgrowthmethodandonlySCanSdatabasetwiceItisaboutanorderofmagnitu

4、defasterthantheApriorialgorithmHowever，itstillencountersperformancebottleneckswhencreatingconditionalFP—treesrecursivelyduringtheminingprocessAdditionallythealgorithmisnotadaptabletodatabaseshavingdifferentcharactersWepr

5、oposetwonovelfrequentpatternminingalgorithms：LIFPGandMIFPGwhicharebasedonpattemfragmentgrowthmethodThetwoalgorithmsadoptpropersearchingstrategiesTheminingtaskisprocesseddirectlyintheoriginallycreatedandcompresseddatastru

6、cture，andnoadditionaldatastructureisneeded，whichimprovesminingefficiencyTheperformancestudyshowsthattheyareaboutfourtimesfasterthanFPGrowthalgorithm，andneedaboutonehalfspaceTheyscalesbetterfordensedatasetsAnefficientalgo

7、rithmforminingmaxpaRemsisproposedbasedonLIFPGAsadepthfirstalgorithm，itminesfrequentpattempossiblylongandadoptsdivideandconquerfilteringmethod，whichgreatlyreducesthesearchspaceTheperformanceshowsthatitisseveraltimesfaster