欢迎来到天天文库
浏览记录
ID:34529778
大小:1.67 MB
页数:69页
时间:2019-03-07
《基于多标准信誉模型地垃圾语音检测》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、华中科技大学硕士学位论文AbstractWiththedevelopmentofcommunicationtechnology,mobilephoneusersareincreasingrapidly.Althoughpeopleenjoytheconvenienceofthistechnology,cellphonegraduallybecomethetargetofspamcallsbecauseofitspersonalandprivacyfeature.Accordingtosurvey,morethan90%peopleareexposedtotheharassmentofspamc
2、alls,thesurfeitofspamcallsareannoyingpeople’sdailylifewhilewastingmuchbandwidthresource.Ontheconditionthatwelackenoughbylawstoimposesanctionstothosespam-call-senders,it’sabetterchoicetousetechnologicalmethodstoweakenthosespamcalls.Therearemanymethodshavebeenproposedtodetectspam-calls,suchas black-an
3、d-whitelist,Turingtest,prepaidfees,thesemethodshavealreadyshownsome effects.Butmoreandmorespam-callsarehumanissue,theycanbypassthemachine detection;theyaretoocunningtodetect.Sopeoplebegintofocusonthereputation system.Theresearchonreputationsystemhavealreadyachievedgoodresults,butthere arestillsomeis
4、sueshaven’tbeentakenintoaccount:1)Howtodealwithmulti-standard criteria?2)Howtoacceleratelearningspeed?3)Howtomakefiltermodelmoreflexible whenusers’behaviorischanging?Tosolvethoseproblemsabove,thispaperproposesamulti-standardreputationmodel basedongrouplearningsolution.Theusertobeclassifiedaccordingt
5、otheirdispositionto trustsolvedthemulti-modelevaluationcriteriaproblem;grouplearningmechanismhas beenestablishedafterdividingusersintodifferentsmallgroups,sothemodelprocessing speedwillnotslowdownbecauseofthelargeamountsofdata;besides,withthestream dataprocessingmethodsandtimingoftheslidingwindowmod
6、elupdatemechanism ensuresthatthemodelisabletochangeaccordingtotheuser'schangestoensurethe accuracyofthemodel.Simulationresultsshowthattheintroductionofdividingtrusttendencymakesthe judgmentresultofthemodel,comparedtoordinaryreputationmodels,havebetter detectioneffect;withtheintroductionofgrouplearni
7、ngmechanism,learningrate improvedsignificantlyevenifeachnodeintheconditionoflowtransactionrates.The modelmeetsthedesignrequirementsverywell.Keywords:spam-call;multi-standardcriteria;learningbasedongro
此文档下载收益归作者所有