TheNewAlgorithmoftheItem-basedonMapReduceZHAOWei1,a1CollegesoftwareTechnologySchool,ZhengzhouUniversityZhengzhou450002,Chinaaiezhaowei@163.
comKeywords:RecommendationsystemparallelcomputingClusteringAbstract.
TraditionalcollaborativefilteringalgorithmbasedonitemandK-meansclusteringalgorithmarestudied,theparallelalgorithmofcollaborativefilteringItem-basedonMapReduceisproposedbyusingMapReduceprogrammingmodel.
Thealgorithmismainlydividedintotwosteps,onestepisK-Meansalgorithmclusteringforusers,anotherstepistheparallelItem-basedalgorithmforclusteringuserrecommendation.
Experimentalresultsshowthatthealgorithmhasobtainedverygoodeffect,improvedtherunningspeedandexecutionefficiency,theimprovedalgorithmismuchsuitableforprocessingbigdata.
IntroductionBigdatausuallyincludesdatasetswithsizesbeyondtheabilityofcommonlyusedsoftwaretoolstocapture,curate,manage,andprocessdatawithinatolerableelapsedtime.
Bigdataishighvolume,highvelocity,and/orhighvarietyinformationassetsthatrequirenewformsofprocessingtoenableenhanceddecisionmaking,insightdiscoveryandprocessoptimization.
Volumemeansbigdatadoesn'tsample;itjustobservesandtrackswhathappens;Velocitymeansbigdataisoftenavailableinreal-time;Varietymeansbigdatadrawsfromtext,images,audio,video;plusitcompletesmissingpiecesthroughdatafusion[1].
Therefore,thebigdatamustbethroughthecomputerstatistics,comparison,analysisofthedatacanbetheobjectiveresults.
Nowelectroniccommercesystemsofeverytransaction,everyinputandeverysearchcanasdata,datathroughthecomputersystemtodothescreening,sorting,analysis,sothattheanalysisresultsisnotonlyanobjectiveconclusion,moreabletohelpbusinessprovidedthedecision-makingofenterprisesandalsocollectedusefuldatacanalsobereasonableplanning,activelyguidethedevelopmentoflargerpowerconsumption,andmoreeffectivemarketingandpromotion.
Withtheincreasingamountofdataintheelectroniccommercesystem,theneedforalargenumberofdatadepthanalysisisincreasinglyurgent.
Therefore,theuseofasimpleandhighscalabilityoftheprogramfortheanalysisofproductrecommendationisparticularlyimportant.
Atpresentdomesticmanyecommercesitesusecollaborativefilteringalgorithm,suchasAmazon,Dangdang,collaborativefilteringalgorithmismainlydividedintobasedontheitemsofthecollaborativefilteringalgorithmanduserbasedcollaborativefilteringalgorithm.
Basedonitemsofcollaborativefilteringalgorithmistomeasurethesimilaritybetweenitemsaccordingtotheuser'spreferences,donotneedtoconsidertheitemspecificcontentfeatures,sothealgorithmismainlyusedine-commercerecommendationandmovierecommendationdomain,thealgorithmwhileinthefieldofelectroniccommercerecommendationhasbeenacertaindegreeofsuccess.
Butinmassivedataarerecommendedwhenthedataisrecommendedperformanceisnothighandthedatainformationlackofsharingandextendedtheleadtothehardwarerequirementscomparedhigherinherentshortcomingsmakeitdidnotreceiveapromotionandsupportofenterpriseelectroniccommerce[2].
SoifweuseMapReducetoachievedistributedparallelcomputing,itwillgreatlyimprovetheefficiencyandperformanceofthealgorithm,andpromotethefurtherdevelopmentofthealgorithm[3-4].
Basedontheitemsofthecollaborativefilteringalgorithmisaccordingtoitemsimilarityanduserhistoryaccessrecordrecommendedtotheusertogeneratealistofitems,buttherearesomesmallproblems,suchasdatasparsityproblemandwhenthemassofusersandthenumberofitems,theuserbehaviorandrecorddatawillgreatly,andthealgorithmforcomputingitemswithsimilarmatrixcostgreatly,algorithmefficiencyandperformancewillgreatlyreduce.
Aimingattheaboveproblems,theclusteringalgorithmhasalsobeenappliedtoacollaborativefilteringalgorithmbasedonitem,themassiveuserclusteringanalysis,soitcanavoidthequestioncarefully,foreachusertorecommendoperation.
Thefirstshoppinguserswithsimilarinterestsintoauserclass,withaclusterofuserrecommendedgoodsarethesame.
Thesecondistoreducethemassiveuserdimensionsbecomedozensofclusteringlimited,thetimecomplexityencounteredabottleneck,andtheparallelclusteringalgorithmusingMapReduceistheeffectivewaytosolvethebottleneck[5].
MapReduceisadistributedprogrammingmodelframeworkonHadoopplatform,intheconditionofnotfamiliarwiththeunderlyingdetailsofthedistributedimplementationoftheimplementationoftheprogram[6].
TheMapReduceasparallelcomputingprogrammingmodel,firstofalltousersofMapReducebasedparallelclusteringandaccordingtotheresultsofuserclustering,ineveryuserclassusingtheMapReduceparallelcollaborativefilteringrecommendation,eventuallygiveusersareasonablepersonalizedcommodityrecommendationlist.
Therunningtimeofdifferentnodesinthequantitativedataiscomparedwiththenewalgorithm.
Theresultsshowthatthedataprocessingperformanceoftheproposedalgorithmisgreatlyimproved.
TheprincipleofMapReduceprogrammingmodelMapReduceisinHadoopplatformbyusingparallelcomputingprogrammingmodel,thetechniqueisproposedbyGoogleforatypicaldistributedparallelprogrammingmodel,theuserintheMapReducemodeldevelopthemapandreducefunctions,canrealizetheparallelprocessing.
Mapwillberesponsiblefordatadispersion,Reduceisresponsiblefordataaggregation.
UsersonlyneedtoachieveMapandReducetwointerface,youcancompletethecalculationofTBleveldata.
BecauseoftheMapReducemodel,thedetailsoftheparallelandfault-tolerantprocessingareencapsulated,whichmakesprogrammingveryeasytoimplement.
MapReduceparallelcalculationisdividedintotwoparts,thefirststepisinitializingtheoriginalinputdatafileandthedatasetisdividedintoapluralityofacertainsizeofdatablock,facilitateparallelcomputing;thesecondstepistostartthemapandreducefunctionsalgorithmofparallelcomputing,finallyproducedthefinalresult.
Figure1ParallelflowchartofMapReduceKeytechnologyresearchandImplementation1.
ThebasicideaofthetraditionalcollaborativefilteringalgorithmbasedonItem-basedThetraditionalbasedonitemsofcollaborativefilteringalgorithmthebasicideaisdividedintothreeparts,thefirstpartistocomputethesimilaritybetweenitems,commonsimilaritycalculationmethodwithcosinesimilarity,Pearsoncorrelationcoefficient,Tanmotocoefficientcorrelationof.
ThispaperselectstheEuclideansimilarityalgorithm,asfollows:TheassumptionisthatthereisavectorXandavectorY:X=(1x,2x,3x),Y=(1y,2y,3y),UsingtheEuclideansimilarityalgorithmtocalculatethesimilaritybetweenXandYSvector(x,y)formulaisasfollows[7]:1(,)1(,)Sxydxy=+(1)Where(,)dxyisthedistancebetweenthevectorXandY,thecalculationformulaisasfollows:222231123(dxyxyyyxx2)Thesecondpartistocalculatetheuserratingsmatrixontheitemsofthegoodsaccordingtothesimilaritymatrix;thethirdpartistheitemsimilaritymatrixWandtheusersoftheitemscorematrixmultiplicationtoobtaintherecommendationresults.
TraditionalItem-Basedcollaborativefilteringrecommendationalgorithmbasedonitemisthestagethataffectstheperformanceofthealgorithm.
Ifthenumberofusersisn,thenumberofcommodityitemsism,thetimecomplexityoffindingalltheitemsinthenprojectisO(2m),thetotalsearchspaceisnusers,sothetimecomplexityofcomputingsimilarityisO(2nm).
Sowhencalculatingthesimilaritymatrixofitems,itisindependentofthesimilaritybetweenthecalculatedandtheotherpairofitemstoaproject,soitispossibletocalculatethesimilaritymatrix.
2.
AnewalgorithmofItem-basedbasedonMapReduceThenewalgorithmismainlydividedintotwosteps;thefirststepistheMapReduceimplementationofK-Meansalgorithmbasedonclusteringofusers.
ThesecondstepistoachievetheparallelrecommendationalgorithmofItem-basedonMapReduce,theproductofuserclusteringrecommendation.
2.
1ThenewalgorithmK-MeansbasedonMapReduceThebasicideaofthetraditionalK-meansclusteringalgorithm:fromMdataobjectsinarbitrarychoiceofKobjectsastheinitialclustercenters;fortherestoftheotherobjects,accordingtotheirdistanceandtheclustercenters,respectively,theyallocatedtoitsmostsimilarclustering;thencalculateeachreceivedanewclusteringalgorithmclusteringcenter;keeprepeatingtheprocessuntilnochangesinacore.
Inthek-meansalgorithmtocalculatethedistancebetweendataobjectsandclustercentersisthemosttime-consumingoperation.
ThedataobjectandKclustercenterdistancecomparisonatthesametime,datafromotherobjectscanalsobecomparedwiththeKdistanceofthecenterofcluster,sotheoperationcanbeparallelized[8]BasedonMapReduceparallelimplementationofK-meansalgorithmcanimprovethespeedoftheclusteringalgorithm,isdividedintothreesteps:thefirststep:themapfunction,foreverypointcalculationrecentlythecenterdistanceandthecorrespondingtothenearestclustercenter.
Thesecondstep:Combinefunction,justcompletedtheMapmachineonthemachinearecompletedwiththesamepointoftheclusterpointofsummation,reducetheamountofcommunicationandcomputationofReduceoperation.
ThisstepisthekeytotheuseofCombinefunctiononthemachineonthefirstofthesameclustermerge,reducedtotheReducefunctionofthetransferandtheamountofcomputation.
Thethirdstep:theReducefunction,theintermediatedataofeachclustercenterwillbeformedandthenewclustercentercanbeobtained.
Eachiterationisrepeatedonthethreestep.
Figure2ParallelFlowChartofK-meansAlgorithmbasedonMapReduce2.
2thecollaborativefilteringalgorithmbasedonMapReduceforparallelimplementationofItem-basedBasedonthesimilaritycalculationformulamentionedabove(1),thispaperpresentsacollaborativefilteringrecommendationalgorithmbasedonMapReduce.
Algorithm1ThecollaborativefilteringrecommendationalgorithmbasedonMapReduceINPUT:Userinformationfile,Iteminformationfile,IntendeduserOUTPUT:IntendeduserrecommendedlistTheprocessisasfollows:Step1:Transformingtheuservectorintoanitemvector;Step2:Parallelcalculationofthesimilaritybetweenitems;thecalculationofthesimilaritybetweenitemsaccordingtotheformula(2)tocalculate;Step3:Similaritymatrixofparallelcomputingobjects;Step4:Parallelcomputinguserratingmatrix;inthecalculationoftheuser'sscoringmatrix,iftheuserisnotontheitemstoomuch,thenthedefaultscoreis1;Step5:Theresultsobtainedbythemultiplicationofthesimilaritymatrixofparallelcomputingobjectsandtheuser'sscorematrixarerecommended.
Experimentalresultanalysis1.
experimentalenvironmentThesimulationexperimentusingVMware_Workstation_10.
0.
3,virtualizationsoftwaretovirtualHadoopcloudplatform.
EightvirtualmachinesareinstalledonthevirtualHadoopcloudplatform,andaHadoopclusterenvironmentisbuiltontheseeightvirtualmachines.
OneofthevirtualmachineasagoodJobTrackernodeNameNode,theothersevenvirtualmachinesdeployedTaskTrackerandDataNode.
Thesemachinesareinthesamelocalareanetwork.
Theexperimentuseseightsetsofvirtualmachinehardwareconfigurationandsoftwareconfigurationasshownintable1:Table1HadoopClusterConfigurationOSCentos6.
4JDKVersion1.
6.
0Hadoop1.
1.
2HardWare2GRAM100GHardDisk2.
ExperimentandanalysisBasedonMapReduceparallelimplementationofItem-basedcollaborativefilteringalgorithminparallelmodeexpansionrateperformancecomparisontest,selectthesizeofthedataset,respectively,intheefficiencyof1-8nodesrunning.
Theexperimentalresultsareshownbelow:Figure3PerformanceTestChartFigure3isbasedonMapReduceparallelimplementationofitembasedcollaborativefilteringalgorithmcantestchart,theXaxisisthenumberofclients,they-axisistheresponsetimeofthesystem.
TheexperimentalresultsshowthatbasedonMapReduceparallelimplementationofitembasedcollaborativefilteringalgorithmperformancecomparedtothetraditionalrecommendationalgorithmissignificantlyimproved.
ConclusionInthispaper,anewalgorithmofcollaborativefilteringalgorithmbasedonMapReduceisproposed.
Theexperimentresultsshowthatthenewalgorithmhashighefficiencyandcanachievehighperformanceatalowcost.
Butinthispaper,theuserclusteringiscompletedonthebasisoftheuserwithasmallnumberofattributes,forhighdimensionalattributesoftheusergroups,butalsotodofurtherresearch.
Inadditiontothenewalgorithminthispaperhasbeenputforward,wewillcontinuetoimprovetheexperimentalmethod,andconstantlyimprovetheaccuracyoftherecommendationalgorithm.
References[1]Chenruming,Challenges,valuesandcopingstrategiesintheeraofbigdata[J].
MobileCommunications.
2012(17):14-15.
[2]SunLingfang,ZhangJing.
ElectronicrecommendationmechanismbasedonRFMmodelandcollaborativefiltering[J].
JournalofJiangsuUniversityofScienceandTechnology(NaturalScienceEdition).
2010,24(3):285-289.
[3]LIGai,PANRong.
etCollaborativefilteringalgorithmparallelizeresearchbasedonlargedatasetsa[J].
ComputerEngineeringandDesign,2012,33(6):2437-2441.
[4]LIWenhai;XUShuren;DesignandimplementationofrecommendationsystemforE-commerceonHadoop[J].
ComputerEngineeringandDesign,2014(35):131-136.
[5]SUNTianhao,LIAnnenget.
ResearchonDistributedCollaborativeFilteringRecommendationAlgorithmBasedonHadoop[J].
ComputerEngineeringandApplications,2014,51(15):124:128[6]XieXuelian,LiLanyou.
ResearchonParallelK-meansAlgorithmBasedonCloundComputingPlatform[J].
ComputerMeasurement&Control,2014,22(5):1510-1512.
[7]YanCun,JiGenlin.
DesignandImplementationofItem-BasedParallelCollaborativeFilteringAlgorithm[J].
JOURNALOFNANJINGNORMALUNIVERSITY(NaturalScienceEdition),2014,37(1):71-75.
[8]WAGNFei,QinXiaolin.
Algorithmfork-meansBasedonDataStreaminCloudComputing[J].
ComputerScience,2015,42(11):235:239.
有在六月份的时候也有分享过新网域名注册商发布的域名促销活动(这里)。这不在九月份发布秋季域名促销活动,有提供年付16元的.COM域名,同时还有5个+的特殊后缀的域名是免费的。对于新网服务商是曾经非常老牌的域名注册商,早年也是有在他们家注册域名的。我们可以看到,如果有针对新用户的可以领到16元的.COM域名。包括还有首年免费的.XYZ、.SHOP、Space等等后缀的域名。除了.COM域名之外的其他...
virmach怎么样?virmach家这几年非常火,从商家的黑五闪购开始,以超低的价格吸引了大批的国人客户,而且商家的机器还是非常稳定的,站长手里的4.75刀年付已经用了两年了,非常稳定,不过商家到国内的线路一般,目前商家新上了夏季优惠促销,价格低到发指,年付7.2美元起,商家反馈将在9月开始更换AMD+NVMe平台,这个消息从年初就有了,不过一直没有更换,目前这个时间也不确定是否准确。点击进入:...
第一、香港服务器机房这里我们可以看到有提供四个大带宽方案,是全向带宽和国际带宽,前者适合除了中国大陆地区的全网地区用户可以用,后者国际带宽适合欧美地区业务。如果我们是需要大陆地区速度CN2优化的,那就需要选择常规的优化带宽方案,参考这里。CPU内存硬盘带宽流量价格选择E3-12308GB240GB SSD50M全向带宽不限999元/月方案选择E3-12308GB240GB SSD100M国际带宽不...
centos6.0为你推荐
哈利波特罗恩升级当爸哈利波特 13年前的晚上发生了什么?硬盘的工作原理硬盘的工作原理?是怎样存取数据的?咏春大师被ko练了十几年的 “ 咏春高手”~~被练一年空手道的轻易打败,难道如今的国术就像国足,不堪一击~~mathplayer如何学好理科老虎数码86年属虎的吉祥数字和求财方向百度关键词分析百度竞价关键词分析需要从哪些数据入手?www.vtigu.com初三了,为什么考试的数学题都那么难,我最多也就135,最后一道选择,填空啊根本没法做,最后几道大题倒m88.comm88.com现在的官方网址是哪个啊 ?m88.com分析软件?45gtv.comLETSCOM是什么牌子?hao.rising.cnIE主页被瑞星绑架http://hao.rising.cn//?b=84主页明明设置的是百度但打开后是瑞星导航,
厦门域名注册 网站域名备案查询 骨干网 韩国空间 鲨鱼机 韩国电信 unsplash java空间 国外在线代理 帽子云 酷番云 东莞服务器 优酷黄金会员账号共享 新睿云 湖南idc 独立主机 电信宽带测速软件 国内空间 葫芦机 开心online 更多