recommendations37

DISCOVERYANDANALYSISOFWEBUSAGEMININGMARATHEDAGADUMITHARAMR.
C.
PatelA.
C.
S.
College,Shirpur,Maharashtra,IndiaABSTRACTInthispaperwedescribesomeofthemostcommontypesofpatterndiscoveryandanalysistechniquesemployedintheWebusagemining.
InthispapermentionAssociationandClusterAnalysis.
AssociationRuleisafundamentalofDataminingtask.
Itsobjectivetofindallco-occurrencerelationshipcalled,Associationamongdataitem.
LetI={i1,i2,…,im}beasetofitems.
LetT=(t1,t2,…,tn)beasetoftransactions.
ClusteranalysisandvisitorssegmentationClusteringisadataminingtechniquethatgroupstogetherasetofitemshavingsimilarcharacteristics.
Intheusagedomain,therearetwokindsofinterestingclustersthatcanbediscovered:userclustersandpageclusters.
GoalDiscoveryandanalysisofwebusagepatternsusingAssociationanalysis.
DiscoveryandanalysisofwebusagepatternsusingClusterAnalysisandVisitorssegmentation.
KEYWORDS:AssociationAnalysis,ClusterAnalysisandVisitorsSegmentationINTRODUCTIONAssociationrulediscoveryandstatisticalcorrelationanalysiscanfindgroupsofitemsorpagesthatarecommonlyaccessedorpurchasedtogether.
AssociationbasedonApriorialgorithm.
Thisalgorithmfindsgroupsofitemusingsupportandconfidence.
Satisfyingauserspecifiedminimumsupportthreshold.
Suchgroupsofitemsarereferredtoasfrequentitemsets&frequentitemsetsgraph.
Logfilesgeneratedbywebserverscontainenormousamountsofwebusagedatathatispotentiallyvaluableforunderstandingthebehaviorofwebsitevisitors.
Clusteringofuserrecords(sessionsortransactions)isoneofthemostcommonlyusedanalysistasksinWebusageminingandWebanalytics.
Clusteringofuserstendstoestablishgroupsofusersexhibitingsimilarbrowsingpatterns.
Suchknowledgeisespeciallyusefulforinferringuserdemographicsinordertoperformmarketsegmentationine-commerceapplicationsorprovidepersonalizedWebcontenttotheuserswithsimilarinterests.
Furtheranalysisofusergroupsbasedontheirdemographicattributes(e.
g.
,age,gender,incomelevel,etc.
)mayleadtothediscoveryofvaluablebusinessintelligence.
Usage-basedclusteringhasalsobeenusedtocreateWeb-based"usercommunities"reflectingsimilarinterestsofgroupsofusers,andtolearnusermodelsthatcanbeusedtoprovidedynamicrecommendationsinWebpersonalizationapplications.
ASSOCIATIONRULESupport&ConfidenceTheSupportofrule,XYthepercentageoftransactioninTthatcontainsXUY.
nisthenumberoftransactioninT.
Supportisusefulmeasurementofitemsetoritems.
IfXistruethenchecksforY,ifXisfalsethennothingtobesayY.
InthefollowingexampleXunionYthencount.
InternationalJournalofComputerScienceEngineeringandInformationTechnologyResearch(IJCSEITR)ISSN2249-6831Vol.
3,Issue1,Mar2013,313-320TJPRCPvt.
Ltd.
314MaratheDagaduMitharame.
g.
(XUY).
CountSupportN(XUY).
CountConfidenceX.
CountUsingaboveexampleswecanaccepttheminsubandminconf.
Tocalculateminsubandminconfasfollows.
T1C++,JAVA,RUBYT2C++,ASPT3ASP,VBT4C++,JAVA,ASPT5C++,JAVA,PHP,ASP,RUBYT6JAVA,PHP,RUBYT7JAVA,RUBY,PHPJAVA,PHPRUBY[sup=3/7,conf=3/3]Inabove7transactionsJAVA,PHP&RUBYshow3/7times.
EveryitemchecksitemsettoeveryusingJoiningandPruningsteps.
Inwebusageminingsuchrulecanbeusetooptimizestructureofwebsite.
e.
g.
Language,/product/softwareRCPACSCOLLEGEWebsiteEXPERIMENT-FINDINGWEBUSAGEASSOCIATIONRULESInstances:14Attributes:5outlooktemperatureDiscoveryandAnalysisofWebUsageMining315humiditywindyplayIfchecksunny,falseyes[sub1/14conf1/1]Thepurposeofthisexperimentwastogivesomeinsightintotheusefulnessofassociationruleswhentheyareappliedtotheweblogdatasetofaneducationinstitutionandothers.
Weexpectedtofindrulesthatcorrelatetowebpagesthatcontaininformationaboutsunny,rainyortemperatureetc.
SupposethisistransactiontableandfindoutFrequentItemsetthen,T1C++,JAVA,RUBYT2C++,ASPT3ASP,VBT4C++,JAVA,ASPT5C++,JAVA,PHP,ASP,RUBYT6JAVA,PHP,RUBYT7JAVA,RUBY,PHPSize1Size2Size3Size4ItemSetSupp.
ItemSetSupp.
ItemSetSupp.
ItemSetSupp.
C++4C++,JAVA3C++,JAVA,RUBY2C++,JAVA,RUBY,ASP1JAVA5C++,RUBY2C++,JAVA,ASP2C++,JAVA,RUBY,PHP1RUBY4C++,ASP3JAVA,RUBY,ASP1ASP4C++,PHP1JAVA,RUBY,PHP3VB1JAVA,RUBY4RUBY,ASP,PHP1PHP3JAVA,ASP2JAVA,PHP3RUBY,ASP1RUBY,PHP3ASP,PHP1Figure1:WebTransactionsandResultingFrequentItemsets(Minsup=1)FindoutFrequentItemsetbyUsingJoiningandPruningMethodsofAssociationRuleFREQUENTITEMSETGRAPHFig.
2,findsitemsC++andRUBYascandidaterecommendations.
TherecommendationscoresofitemAandCare1,correspondingtotheconfidencesoftherules,JAVA,ASP->C++andJAVA,ASP->RUBY,respectively.
Aproblemwithusingasingleglobalminimumsupportthresholdinassociationruleminingisthatthediscoveredpatternswillnotinclude"rare"butimportantitemswhichmaynotoccurfrequentlyinthetransactiondata.
316MaratheDagaduMitharamC=C++J=JAVAA=ASPR=RUBYP=PHPFigure2:FrequentItemsetsCLUSTERANALYSISANDVISITORSSEGMENTATIONConceptandExampleClusteringofuserrecords(sessionsortransactions)isoneofthemostcommonlyusedanalysistasksinWebusageminingandWebanalytics.
Clusteringofuserstendstoestablishgroupsofusersexhibitingsimilarbrowsingpatterns.
Suchknowledgeisespeciallyusefulforinferringuserdemographicsinordertoperformmarketsegmentationine-commerceapplicationsorprovidepersonalizedWebcontenttotheuserswithsimilarinterests.
DiscoveryandAnalysisofWebUsageMining317HereweUsetheformulaof"WebDataMining"-Bingliubook.
Asanexample,considerthetransactiondatadepictedinsimplicityweassumethatfeature(pageview)weightsineachtransactionvectorarebinary(incontrasttoweightsbasedonafunctionofpageviewduration).
Weassumethatthedatahasalreadybeenclusteredusingastandardclusteringalgorithmsuchask-means,resultinginthreeclustersofusertransactions.
Itshowstheaggregateprofilecorrespondingtocluster1.
Asindicatedbythepageviewweights,pageviewsBandFarethemostsignificantpagescharacterizingthecommoninterestsofusersinthissegment.
PageviewC,however,onlyappearsinonetransactionandmightberemovedgivenafilteringthresholdgreaterthan0.
25.
Suchpatternsareusefulforcharacterizinguserorcustomersegments.
Thisexample,forinstance,indicatesthattheresultingusersegmentisclearlyinterestedinitemsBandFandtoalesserdegreeinitemA.
GivenanewuserwhoshowsinterestinitemsAandB,thispatternmaybeusedtoinferthattheusermightbelongtothissegmentand,therefore,wemightrecommenditemFtothatuser.
ExperimentandResultsInthisexperimentwedefinetable"weather"anddefinefields.
318MaratheDagaduMitharamOutputUsingClusterinWeka===Runinformation===Scheme:weka.
clusterers.
HierarchicalClusterer-N2-LSINGLE-P-A"weka.
core.
EuclideanDistance-Rfirst-last"Relation:weatherInstances:13Attributes:5outlooktemperaturehumiditywindyIgnoredplayTestmode:Classestoclustersevaluationontrainingdata===Modelandevaluationontrainingset===Cluster0((((((1.
0:0.
18505,1.
0:0.
18505):0.
05959,1.
0:0.
24464):0.
7557,(1.
0:0.
16832,(1.
0:0.
08235,1.
0:0.
08235):0.
08597):0.
83201):0.
00109,((0.
0:0.
22986,0.
0:0.
22986):0.
77157,0.
0:1.
00142):0):0.
00106,(0.
0:0.
21648,0.
0:0.
21648):0.
78601):0.
00135,1.
0:1.
00384)ClusteredInstances012(92%)11(8%)Classattribute:playClassestoClusters:01<--assignedtocluster71|yes50|noCluster0<--yesCluster1<--NoclassIncorrectlyclusteredinstances:6.
046.
1538%DiscoveryandAnalysisofWebUsageMining319VisualizationsofPatternsCONCLUSIONSUsagepatternsdiscoveredthroughWebusageminingareeffectiveincapturingitem-to-itemanduser-to-userrelationshipsandsimilaritiesatthelevelofusersessions.
Thispaperhasattemptedtoforthepurposeofwebusagemining.
TheproposedmethodsweresuccessfullytestedonthedatasetordatabasesusingassociationruleandclusteranalysismethodusingWekaTool.
Ourexperimentsconfirmedthatoneofthemajorissuesinassociationruleandclusterfindingistheexistenceoftoomanyrulesandgroups,allofwhichsatisfydefinedconstraints.
REFERENCES1.
Webdatamining–BingLiu320MaratheDagaduMitharam2.
PPTforWebusagemining-BingLiu3.
Srivastava,J.
,Cooley,R.
,Deshpande,M.
,Tan,P.
N.
(2000).
WebUsageMining:DiscoveryandApplicationsofUsagePatternsfromWebData.
ACMSIGKDD,Jan2000.
4.
JaideepSrivastavaPaper5.
WCA.
Webcharacterizationterminology&definitions.
6.
http://www.
w3.
org/1999/05/WCA-terms/.
Vigenteal19/11/2005