为什么手机通讯录里会无故多很多srhaoma_百度知道
为什么手机通讯录里会无故多很多srhaoma
我有更好的答案
按默认排序
或者不要在用360的拦截功能。，并且使用了骚扰电话拦截功能。把这个几个联系人删掉就行了，如果使用了该功能，360就会下载上千个骚扰电话到你的手机 命名为srhaoma0----srhaoma4，不然还会下载的是不是手机装360
是不是你手机所登陆的小米账号是和朋友或家人的同一个账号，而家人的号码自动同步给你了？
其他类似问题
手机通讯录的相关知识
为您推荐：
等待您来回答
下载知道APP
随时随地咨询
出门在外也不愁手机看此贴：
用得意APP扫一扫
微信扫一扫 转发至朋友圈
UID419697&好友&帖子&主题&精华0&小红花1 &积分93&在线时间0 小时&注册时间&阅读权限25&最后登录&
UID419697&帖子&精华0&金币93 &威望0 &注册时间&
可是打过来的电话号码确实就是那个srhaoma里面的咧！哎，不清楚，烦死了
UID419697&好友&帖子&主题&精华0&小红花1 &积分93&在线时间0 小时&注册时间&阅读权限25&最后登录&
UID419697&帖子&精华0&金币93 &威望0 &注册时间&
为了一个这样的变态换号码不值得吧！我号码用了十几年了
UID528906&好友&帖子&主题&精华0&小红花55 &积分1767&在线时间13 小时&注册时间&阅读权限65&最后登录&
UID528906&帖子&精华0&金币1741 &威望0 &注册时间&
UID598776&好友&帖子&主题&精华0&小红花5 &积分1101&在线时间0 小时&注册时间&阅读权限55&最后登录&
UID598776&帖子&精华0&金币1101 &威望0 &注册时间&
一搞有400的骚扰电话捏。见怪不怪了。只能无视
UID419697&好友&帖子&主题&精华0&小红花1 &积分93&在线时间0 小时&注册时间&阅读权限25&最后登录&
UID419697&帖子&精华0&金币93 &威望0 &注册时间&
UID419697&好友&帖子&主题&精华0&小红花1 &积分93&在线时间0 小时&注册时间&阅读权限25&最后登录&
UID419697&帖子&精华0&金币93 &威望0 &注册时间&
我这个骚扰电话来自同一个人，他大概是用一个软件打的，每天打的号码都不一样！
UID528906&好友&帖子&主题&精华0&小红花55 &积分1767&在线时间13 小时&注册时间&阅读权限65&最后登录&
UID528906&帖子&精华0&金币1741 &威望0 &注册时间&
UID198260&好友&帖子&主题&精华0&小红花226 &积分6343&在线时间958 小时&注册时间&阅读权限90&最后登录&
UID198260&帖子&精华0&金币4427 &威望0 &注册时间&
陌生号，我都不接。
UID602609&好友&帖子&主题&精华0&小红花65 &积分1375&在线时间0 小时&注册时间&阅读权限60&最后登录&
UID602609&帖子&精华0&金币1375 &威望0 &注册时间&
貌似苹果不用下360的，我一个做软件的朋友就告诉我，360恶意盗取个人信息呢！
UID419697&好友&帖子&主题&精华0&小红花1 &积分93&在线时间0 小时&注册时间&阅读权限25&最后登录&
UID419697&帖子&精华0&金币93 &威望0 &注册时间&
嗯嗯！以后就只能这样！
UID529080&好友&帖子&主题&精华0&小红花9 &积分1496&在线时间2 小时&注册时间&阅读权限60&最后登录&
UID529080&帖子&精华0&金币1492 &威望0 &注册时间&
UID621913&好友&帖子&主题&精华0&小红花0 &积分498&在线时间0 小时&注册时间&阅读权限40&最后登录&
UID621913&帖子&精华0&金币498 &威望0 &注册时间&
对呀，问度娘
UID154196&好友&帖子&主题&精华0&小红花207 &积分8268&在线时间2723 小时&注册时间&阅读权限100&最后登录&
UID154196&帖子&精华0&金币2782 &威望8 &注册时间&
设置拒绝陌生来电
UID440020&好友&帖子&主题&精华0&小红花22 &积分834&在线时间213 小时&注册时间&阅读权限50&最后登录&
UID440020&帖子&精华0&金币408 &威望0 &注册时间&
换安卓手机试试
UID608099&好友&帖子&主题&精华0&小红花179 &积分1647&在线时间0 小时&注册时间&阅读权限65&最后登录&
UID608099&帖子&精华0&金币1647 &威望0 &注册时间&
你这个是因为下了360才出现的电话号码！苹果手机不能装360！
UID326926&好友&帖子&主题&精华0&小红花21 &积分1847&在线时间525 小时&注册时间&阅读权限65&最后登录&
UID326926&帖子&精华0&金币797 &威望0 &注册时间&
我没有遇到过阿,
也一直用的苹果,冒下过360.冒越狱过..
理想很丰满，现实很骨感。
UID122375&好友&帖子&主题&精华0&小红花191 &积分5832&在线时间1751 小时&注册时间&阅读权限90&最后登录&
UID122375&帖子&精华0&金币2320 &威望2 &注册时间&
苹果手机用360，楼主，知道什么叫脱裤子放屁么
UID279746&好友&帖子&主题&精华0&小红花21 &积分3732&在线时间14 小时&注册时间&阅读权限85&最后登录&
UID279746&帖子&精华0&金币3704 &威望0 &注册时间&
UID419697&好友&帖子&主题&精华0&小红花1 &积分93&在线时间0 小时&注册时间&阅读权限25&最后登录&
UID419697&帖子&精华0&金币93 &威望0 &注册时间&
你这么有文化怎么回答不了别人滴问题咧！文明点不行？
UID419697&好友&帖子&主题&精华0&小红花1 &积分93&在线时间0 小时&注册时间&阅读权限25&最后登录&
UID419697&帖子&精华0&金币93 &威望0 &注册时间&
好吧，谢谢
版权所有& Powered by六一儿童节后的本周应该是欢乐的一周，但是对于我大天朝来说却是沉痛的一周……多说无...
OS X 可扩展的地方还很多，你们还希望 OS X 10.11 能有什么样的更新呢？在 6 月 9 日...
谷歌在发布 Nexus 6 半年之后终于迎来第一次价格调整，谷歌官方商店今天下调了 Nexus ...
瑞士的传统钟表业已经存在几百年的时间了，这一次会不会因为Apple Watch的到来而受到...
对咱们锋友来说，最近有两个日子要逼近了，一个是高考的日子，接着就是苹果一年一度的...
近日 HTC 在台湾举行股东会，会上董事长王雪红为 HTC 的低迷向全体股东鞠躬致歉。HTC ...
今天关于越狱的新闻真不少，太极已经开始瞄准 iOS 9 的越狱了。
如果 iOS 8.3 越狱放出的话，你还关心 iOS 8.4 的推送吗？
一款比较有爱的国产 App《东家—指尖上的中国》，是专门针对国内手工匠人提供的社区平...
《飞天滑板高手（HoverRider）》是一款画面清新难度感人的无限跑酷游戏，游戏全程采用...
蹦床是大家小时候最爱玩的一个项目，而蹦床在游戏中所扮演的角色自然也就是给大家带来...
普通球迷可能很少有机会亲身去体验Tiki Taka战术，不过通过近日在苹果商店当中上架的...
如果说在一两年前，各种 FF 炒冷饭让粉丝们无奈和厌烦，那么现在，SE 就要靠着移动平...
《4x4 Jam HD》是一款越野竞速游戏，造型粗犷的越野车以及遍布世界各地的比赛场地令人...
Ben10是一部科幻动画片，讲述一位10岁小孩田小班Omnitrix偶然发现神秘的高科技外星装...
“Dragon Guardian”手机保护壳，宫廷城堡风、24K 金外加闪亮镶钻……集各种奢华于一...
本周三，D-EYE公司宣布准备推出一款同名的基于iPhone的视网膜成像配件。
如果真有12.9英寸的iPad Pro，外设厂商又要再改进他们的产品了。
如今手机已经占领世界，人们几乎每一刻都离不开手机，对于需要频繁使用手机的人来说，...
苹果估计会少赚 40 亿。
这两家公司的“麻烦事”，究竟何时才是个尽头。
《复仇者联盟2：奥创纪元》上映之后获得了不俗的票房成绩，这也使得越来越多的周边厂...
上周苹果公司推出了全新的iPhone基座产品，兼容性挺高的。
新到手的iphone6 最近联系人里面怎么有360的标志
注册时间 最后登录
在线时间309 小时 UID
主题帖子人气
红苹果, 积分 296, 距离下一级还需 204 积分
本帖最后由 洛马克 于
16:40 编辑
如图所示，刚激活的手机 什么软件都没安装&&可是最近联系人里面有个这个，通讯录里面也有这个名为“SRhaoma”的 里面应该是一堆黑名单 但头像却是360的logo&&让人怀疑是是不是直接被人用过的？ 但是查序列号保修到期是到号 而我是今天5月8号拿到手机的 按查询结果来说应该不会是翻新机 二手机之类的吧？那么问题来了，谁能帮我解释一下 这个360头像的联系人是怎么回事呢？ 系统是ios8.2 还没有插sim卡
注册时间 最后登录
在线时间14 小时 UID
主题帖子人气
网上的答案是不是手机装360，并且使用了骚扰电话拦截功能，如果使用了该功能，360就会下载上千个骚扰电话到你的手机 命名为srhaoma0----srhaoma4。把这个几个联系人删掉就行了，或者不要在用360的拦截功能，不然还会下载的。。希望能帮到你PS：我装了360也是这样
注册时间 最后登录
在线时间0 小时 UID
主题帖子人气
希望能和机友成为好朋友！
注册时间 最后登录
在线时间309 小时 UID
主题帖子人气
网上的答案是不是手机装360，并且使用了骚扰电话拦截功能，如果使用了该功能，360就会下载上千个骚扰电话到 ...
问题是这是刚激活的手机 什么软件都没安装啊？
注册时间 最后登录
在线时间933 小时 UID
主题帖子人气
要么就是自动备份的icould 你激活过之后自动下载到新手机上了
注册时间 最后登录
在线时间309 小时 UID
主题帖子人气
要么就是自动备份的icould 你激活过之后自动下载到新手机上了
确实是登陆了icloud&&你这么说 我估计也有这个可能& &但是以前的几个手机上都没有下载过360啊 为什么会有360的logo
注册时间 最后登录
在线时间50 小时 UID
主题帖子人气
要么就是自动备份的icould 你激活过之后自动下载到新手机上了
注册时间 最后登录
在线时间223 小时 UID
主题帖子人气
icloud备份的数据。。
注册时间 最后登录
在线时间14 小时 UID
主题帖子人气
应该是你备份了
注册时间 最后登录
在线时间89 小时 UID
主题帖子人气
机子如果串号没问题，那保修就没问题，那多出来这个东西肯定就不是商家给你的机子有问题了，应该就是你自己什么时候装了或者别人装的你不知道而已
威锋旗下产品
Hi~我是威威！
沪ICP备号-1 丨 深公安网监备案号 5
增值电信业务经营许可证：
Powered by Discuz!92Learning Personal Social Latent Factor Model for Social Recommendation-第2页
上亿文档资料，等你来发现
92Learning Personal Social Latent Factor Model for Social Recommendation-2
Zi,m,n；Giventhis,theexpectation；andZi,m,n:；E(Zi,m)=；KXn=1；+?；E(Zi,m,n)+E(Zi,m,n)；(22)；Notethatinthisapproach,t；Zi,?isO(TEK2)(ateachE-st；UpdateB:Inaddition,wecan；B(m,n)=P；ui；ui(1；+?
+Zi,m,nGiventhis,theexpectationofZi,mcanbecalculatedbasedon?andZi,m,n:E(Zi,m)=KXn=1+?E(Zi,m,n)+E(Zi,m,n)(22)Notethatinthisapproach,thetimecomplexityforestimatingZi,?isO(TEK2)(ateachE-step),whereTisthenumberoftheclusters.UpdateB:Inaddition,wecanutilizethismethodtoinferBintheM-step:PB(m,n)=Puiui(1+??ρ)(E(Zi,m,n)+E(Zi,m,n))+E(Zi,m,n)(23)whereρisthesparsityparametertocapturenon-interactionofthesocialnetwork[2].Andρcanbeestimatedby:ρ=?m,nE(Zi,m,n)PP+?uim,n(E(Zi,m,n)+E(Zi,m,n))uiPP(24)4.3OverallEMAlgorithmAlgorithm1EM-PSFL-Inference(R,G,Φ)Parameter:R:user-G:Φ:1:RandomInitializeparametersΘ=(U,V,Π,M,B).2:repeat3:ScalableE-Step:4:OnlineK-MeansClusteringonΘ.+?5:EstimateE(Zi,k,l),E(Zi,k,l)andE(Zi,k)usingEqn.19,20and226:M-Step:7:UpdateparametersU,V,MusingEqn.15,16and178:UpdateΠaccordingEqn.13and14.9:UpdateBusingEqn.2310:untilRepeattheE-MstepsuntilΘconvergence.TheparametersinferencemethodforPSFLisoutlinedinAl-gorithm1.ThetimecomplexityfortheE-stepandM-stepareO(TEK2)andO(RKN)4,respectively.IntheAlgorithm1,weusetheonlineK-MeansclusteringalgorithmtoobtainC,whichleadstoO(KTN)timecomplexity.Inpractice,thenum-berofclustersThasarelativesmallaffecttoourproposedPSFLmodel(weuse300inthiswork).Forthehyper-parametersΦinPSFLmodel,theyaredeterminedbycross-validation.Wewillfurtherelaborateitintheexperimentsection.MMMFmaximum-marginmatrixfactorizationmodel(MMMF)placesalatentfactorvectoroneachuseranditem.Inthetrainingstep,user-itemlatentfactorsarecomputedbyuser-ratingmatrixdecomposition.Forprediction,ratingscoresiscalculatedastheinnerproductbetweenuserlatentfactorsanditemlatentfactors.WeimplementMMMFasdescribedin[14].Inaddition,wealsoincorporatetheitem-neighborhoodinformationbasedon[9].InMMMF,therearethreehyper-parameters,regularizationtermsσu,σvonfactorsUandVrespectivelyandσsitemsimilarityconstraintterms.Thethreehyper-parametersaredeterminedbycross-validationandtheresultsreportedinallexperimentalre-sultsarebasedontheparametercon?gurationswhichproducethebestresults.FIPFriend-InterestPropagation(FIP)model[18]utilizesthelatentfactorsofusersnotonlyforpredictingtheratings,butalsoforlinkprediction.Theinnerproductbetweentwouserlatentfactorscandirectlyhelpwhethertheyarelinkedinsocialnet-work.NotethattheoriginalFIPmodelmainlytargetsfortheitemrecommendation(ratingisonlybinary).Forcomparisonpurpose,weextendtheirapproachtotheratingprediction,asinEqn.8.Fourhyper-parametersinthemodel,λs,λu,λv(the?rstthreearethesameastheMMMFmodel),andλEarealldeterminedbycross-validation.SR-LFMSocialregularizationbasedlatentfactormodel(SR-LFM)[11]utilizesthesocialrelationshipsexplicitlytoregu-latethelatentfactsofusers.Speci?cally,theindividual-basedregulationismodeledasinEqn7.Forcomparisonpurpose,wealsoaddneighborhooditemsimilarityconstrainttotheSR-LFMmodel.Therearefourparameters,λs,λu,λv(the?rstthreearethesameastheMMMFmodel),andλE,aredeterminedbycross-validation.PSLFJointPersonalandSociallatentFactormodel(PSLF)proposedinthepaper,utilizesbothusers’pastbehaviorsandso-cialrelationshipsforsocialrecommendation.InthePSLFmodel,therearesixhyper-parameters:λs,λuandλvarethesameasMMMFβisapriorDirichletparameter(assignedasa?xedvaluei.e.1.0);andσMandσ.Alltheseparametersexceptβaredeterminedbycross-validation.Intheexperimentalstudy,weusethetwopopularmetrics,MeanAbsoluteError(MAE)andRootMeanSquareError(RMSE)tomeasurethepredictionqualityofourproposedPSLFmodelincomparisonwithothercollaborative?lteringandsocialrecom-mendationmethods.MAEisde?nedas:1X??MAE=|Rij?Rij|(25)T(ui,vj)5.EXPERIMENTSInthissection,weusetherealworlduserratingdataandtheircorrespondingsocialnetworktoempiricallyvalidatetheeffec-tivenessoftheproposedjointpersonalandsociallatentfactor(PSLF)modelforsocialrecommendation.Wecompareitwiththeexistingsocialrecommendationmethodsandthestate-of-the-artcollaborative?lteringapproacheswithoutconsideringtheso-cialrelationships.Speci?cally,thefollowing?verecommenda-tionmodelsarecompared:SIMitem-orientedneighborhoodmodel(SIM)istheclassicalcollaborative?lteringmethodbasedonitemsimilarity.SIMpre-dictstheratingscoreoftheitemaccordingtotheuserhistoricalratingsonthesimilaritems.WeimplementtheSIMasdescribedin[15].4RMSEisde?nedas:vuu1X??2RMSE=t(Rij?Rij)T(ui,vj)(26)whereRijdenotestheratingscoreuseruigavetomovievj,denotesthepredictedratingscoreuseruigavetomovievj,andTdenotesthenumberoftestedratings.ThesmallerMAEorRMSEvaluemeansabetterperformance.??Rij5.1ExperimentalDatasetsFlixsterDataset.Flixsterisasocialnetworkingserviceinwhichuserscanratemovies5.Userscanaddotheruserstotheirfriendlistandcreateasocialnetwork.ThesocialrelationsThe?ixsterdatasetispublic,anditcanbedownloadfromhttp://www.sfu.ca/sja25/datasets/5NisthedimensionoflatentinterestfactorsTable1:GeneralStatisticofFlixsterandDouBanStatisticsFlixsterDouBanUsers786,SocialRelations7,058,Ratings8,196,Items48,Table3:PerformanceComparisonwithMAEandRMSEon?ixsterdatasetModelMAERMSESIM0.MMMF0.1429±(0.6±(0.0007)FIP0.1400±(0.3±(0.0009)SR-LFM0.1389±(0.4±(0.0017)PSLF0.1339±(0.1±(0.0013)datasetDouban?ixsterTable2:Pairedt-Test(2-tail)resultst-TestMMMFFIPPSLF(MAE)3.38e?83.16e?7PSLF(RMSE)5.73e?92.22e?6PSLF(MAE)7.47e?41.06e?3PSLF(RMSE)1.22e?41.08e?4SR-LFM2.42e?46.52e?73.09e?41.12e?3inFlixsterareundirected.ItalsocontainsratingsexpressedbyusersintheperiodfromNovember2005toNovember2009.PossibleratingvaluesinFlixsterdatasetare10discretenum-bersintherange[0.5,5]withstepsize0.5.Therearetotally786,936users,and48,794moviesinthisdataset.Theover-allnumberofsocialrelationsandratinghistoricalrecordsare7,058,819and8,196,077,respectively.DoubanDataset.DoubanisaChinesesocialwebsiteprovid-inguserrating,reviewandrecommendationservicesformovie,booksandmusic6.ItprovidesFacebook-likesocialnetwork-ingservicesthatuserscanmakefriendswitheachotherthroughtheemailcommunication.DoubandatasetiscrawledandsharedwithusbyMaHao[11].Inthedataset,userscanassign5in-tegerratings(1to5)tomovies,booksandmusics.Thereareoverall129,490uniqueusersand58,541uniqueitems.Thenumberofsocialrelationsandratingrecordsare1,692,950and8,415,420respectively.ThebasicstatisticsoftheFlixsterandDoubandatasetareshowninTable5.1.Table4:PerformanceComparisonwithMAEandRMSEondoubandatasetModelMAERMSESIM0.MMMF0.1427±(0.5±(0.0010)FIP0.1397±(0.6±(0.0010)SR-LFM0.1367±(0.8±(0.0012)PSLF0.1326±(0.4±(0.0013)5.2ExperimentResultsInthissubsection,wereporttheperformanceofdifferent(so-cial)recommendationapproaches.Theusers’ratingdatasetisrandomlydividedinto2folds.Intheexperiments,weuse70percentoftheratingdatafortrainingandtheother30percentfortesting.Therandomselectionwascarriedout5timesinde-pendently,andwereporttheaverageresults.NotethatexceptSIM,theotherfourmodelsMMMF,FIP,SR-LFMandPSLFneedselectthedimensionofthelatent(personal)factorswhichisdenotedbyN.Thus,weshowtheperformanceofMMMF,FIPSR-LFMandPSLFmodelswithdifferentN,whichare(4,6,8,10,20)respectively.ForPSLFmodel,additionalsocialfactordimensionKissettobeK=50.Later,wewillfurtheranalyzetheperformanceofPSLFwhenKvaries.RatingPredictionVaryingN(NumberofLatentFactors):First,weevaluatethe?vedifferentmethodsbyvaryingthenum-beroflatentfactorsN.BothMAEandRMSEmetricsareusedtomeasurethepredictionquality.AsshowninFigure4,theproposedPSLFmodelconsistentlyoutperformstheotherfourmethods.TheperformanceofSIMisgenerallytheworstamongallthesemethodsinbothDoubanand?ixsterdataset.Thisisasexpectedbecauseitdoesnotutilizeanylatentfactors(astraightlineinthese?gures).TheMMMFisbetterthanSIMasituti-lizesthelatentfactorsanditemneighborhoodinformation.WeobservethethreerecommendationapproachesFIP,SR-LFM,andourPSLFareallconsistentlybetterthanMMMFandSIM.This6providesastrongevidencethatthesocialrecommendationin-deedisusefulandcanbeusedtoimprovetherecommendationaccuracy.Inthesesocialrecommendationapproaches,SR-LFMisslightlybetterthanFIPinDoubandataset,andcomparablewithFIPin?andourPSLFarealwaysbetterthanbothofthem.FromtheFigure4,weobservethataconvergenceeffectofN:whenweincreasethelatentfactordimensionNtobearound10,thereseemtobelittleimprovementforanylargeN.Thissug-geststhatasmallnumberoflatentfactors(suchas10)isenoughforallthemodels,includingMMMF,FIP,SR-LFM,andPSLF.DetailedPerformanceAnalysisforN=10InTable5.2and5.2,weprovideadetailedcomparisonofthese?veapproaches(forthefourlatentfactorbasedapproaches,thenumberoflatentfactorNissettobe10.)WeobserveintermsofMAEin?ixsterdataset,PSLFmodelcanimprovetheperformanceashighas3.60%incontrasttoSR-LFMmodel,4.36%and6.30%incon-trasttoFIPandMMMFmodels,respectively.IntermsofRMSEin?ixsterdataset,PSLFmodelimprovestheperformanceashighas1.39%,1.71%and2.91%incontrasttoSR-LFM,FIP,andMMMF,respectively.InDoubandataset,PSLFmodelimprovestheperformanceashighas2.99%,5.06%,7.07%intermsofMAE,and1.39%,1.85%,2.91%intermsofRMSEincontrasttotheotherthreelatentfactorbasedapproaches,suchasSR-LFM,FIP,andMMMF,respectively.Finally,tovalidatethestatisticalsigni?canceofourexperi-ments,weperformthepairedt-test(2-tail)overtheMAEandRMSEoftheexperientialresult.AsshowninTable5.1,allthet-testresultsarelessthan0.01,whichmeanstheimprovementsofPSLFoverothermethodsarestatisticallysigni?cant.ParameterSensitivityAnalysis:Here,weanalyzetheperfor-manceofPSLFmodelwhenvaryingthedimensionofsocialfactorsK.Intheexperiments,wecomparetheperformanceonMAEwithdifferentsocialfactordimensions,whichare(20,50,80).AsshownintheFigure5,weobservethattheperformancewithK=50andtheperformancewithK=80theyarebothslighterbetterthanK=20.Wenotethatinmostofthecases,theperformancewithK=50isevenslightlybetterthantheperformancewithK=80.Thisindicatesalargenumberofsocialfactors(beyond50)maynothelpmuchforperformanceimprovement.(b)RMSE(Douban)(c)MAE(?ixster)(a)?ixster(b)DoubanFigure6:Correlationbetweenuser’sratingbehav-iorsandsocialactivitiessamplingmethod(randompair),and0.0561basedonsec-samplingmethod(randomfriends)withnearly7percentim-InDoubandataset,theaverageinterestcoef?cients0.2554and0.2654basedonrandomsamplingmethodandsamplingmethod,respectively.Thereis4percentim-forsocial-awaresamplingmethod.Therefore,simplyusershavesimilarinterestwiththeirfriendsinsocialmaystillbehelpfulwhenincorporatingintorecom-system,thoughtheimprovementmaybelimited.wecomparethedistributionofPearsoncorrelationco-cient(PCC)betweenthetwosamplesetsgeneratebythetwoapproaches.First,PCCissortedinascendingandde-orderrespectively.Foreachorder,wegetTopNper-ofsamples,andcalculatetheaveragePCC.InFigureweshowthecomparisonwithTop(N)averagePCCinas-orderonrandomsamplingsetandfriendsamplingset.theFigure5.3,wecanobservethatthefriendsamplingPCCcurveisaboverandomsamplingmethod’sinbothanddoubandataset.Thisindicatesthatfriendsinsocialhavelessprobabilitytohaveoppositeinterests.InFig-5.3,wereportthecomparisonwithTop(N)averagePCCinorderbetweenthetwosamplingapproaches.Here,foundthatthefriendsamplingmethod’sPCCcurveisbe-randomsamplingmethod’s.Itindicatesthattherearemanywhohavesimilarinterestsmaynotbefriendswitheachinsocialnetwork.Thisagainsuggeststhatusingthe(per-latentfactortomeasurethelikelihoodofbeingfriendsisaccurate.??7(a)?ixster(b)Douban7:ComparisonwithTop(N)PCCindescendingorderRELATEDWORKstudiesshowusers’trustrelationshipcanbeemployedtraditionalrecommendersystems[13].Afewtrust-recommendationmethodshavebeenproposed[7][4].How-8.ACKNOWLEDGMENTSTheauthorsappreciatetheanonymousreviewersfortheirex-tensiveandinformativecommentstohelpimprovethepaper.WealsothankDrHaoMaforsharingDoubanDataset.Ye-longShenandRuomingJin’sworkinthispaperispartiallysup-portedbyNationalScienceFoundationunderCAREERAwardIIS-0953950.(a)?ixster(b)DoubanFigure8:ComparisonwithTop(N)PCCinascendingorderever,trust-basedrecommendationapproachesaredifferentfromsocialrecommendationinmanyaspects.Typically,ontheweb-site,whenauserAlikesareviewbyanotheruserB,userAprobablywilladduserBtohis/hertrustlist[11].ThisprocessoftrustgenerationisaunilateralactionthatdoesnotrequireuserAtocon?rmtherelationship.However,thesocialrelationshipreferstothecooperativeandmutualrelationshipsthatsurroundus,suchasclassmates,colleagues,orrelatives,etc.Inaddition,trust-basedrecommendersystemsarebasedontheassumptionthatusershavesimilartasteswithotheruserstheytrust.Thishypothesismaynotalwaysbetrueinsocialrecommendationsys-temssincethetastesamongtheuserandher/hisconnectedusersmayvarysigni?cantly.Amoredetailedanalysisonthediffer-encebetweenthetrust-basedrecommendationapproachandthesocialrecommendationapproachcanbefoundin[11].Therehavebeenafewattemptsonsocialrecommendationslately[5,11,18],whichareallbasedontheassumptionthatanypairoffriendsinthesocialnetworkshallhavesimilarinterests.Therecentstudies[11,18]incorporatesuchanetwork-basedsimilaritypropertybetweenusersintothestate-of-the-artmatrixfactorizationrecommendationapproaches[10].However,theyalmostcompletelyignoredtheheterogeneityanddiversityofthesocialrelationship.Socialattribute(factor)extractionwas?rstproposedbyHol-landetal.[8]andHarrisonetal.[17].Theyarealsoreferredtoastheblockmodel,whichcanbeusedtodescribetheroleorpositionofactorsinthesocialnetwork.In[8]and[17],usersinsocialnetworkhaveonlyonefactor(position)beingextracted,andtheinteractionofthepositionsiscapturedinthelatentso-cialstructures.Recently,morecomplicatedblockmodel,mixed-membershipblockmodel[2],proposedtorelaxtheassumptionofsinglepositionbloackmodel,allowuserstohaveamorenat-ural,mixed-membershipofsocialattributes.However,theexit-ingparameterinferencemethodhasscalabilitybottleneck,whichrequiresO(n2)timecomplexityineachiteration,wherenistheusernumberofsocialnetworks.Inthiswork,wedevelopascaleEMapproachwithtimecomplexityO(E)ineachiteration,whereEisthenumberofedgesinthesocialnetwork.9.REFERENCES[1]GediminasAdomaviciusandAlexanderTuzhilin.Towardthenextgenerationofrecommendersystems:Asurveyofthestate-of-the-artandpossibleextensions.IEEETrans.onKnowl.andDataEng.,17:734C749,June2005.[2]EdoardoM.Airoldi,DavidM.Blei,StephenE.Fienberg,andEircP.Xing.Mixedmembershipstochasticblockmodels.InJMLR’08.TheJournalofMachineLearningResearch,pages.ACM,2008.[3]PaulH.CalamaiandJorgeJ.More:9A.Projectedgradientmethodsforlinearlyconstrainedproblems.Math.Program.,39:93C116,October1987.[4]J.O’DonovanandB.Smyth.Trustinrecommendersystems.InIUI’05.Proceedingsofthe10thinternationalconferenceonIntelligentUserInterfaces,pages167C174.ACM,2005.[5]H.J.LeeF.Liu.Useofsocialnetworkinformationtoenhacecollaborative?lteringperformance.InESA’10.ExpertSystemswithApplications,pages.ElsevierLtd,2010.[6]QuanquanGuandJieZhou.neighborhoodpreservingnonnegativematrixfactorization.InBMVC,pages1C11,2009.[7]MichaelR.LyuHaoMa,IrwinKing.Learningtorecommendwithsocialtrustensemble.InSIGIR’09.Proceedingsofthe32thInternationalConferenceoninformationretrieval,pages203C210.ACM,2009.[8]PWHolland,KBLaskey,andSLeinhardt.Stochasticblockmodels:Firststeps.SocialNetworks,5(2):109C137,1983.[9]YehudaKoren.Factorizationmeetstheneighborhood:amultifacetedcollaborative?lteringmodel.InKDD’08.Proceedingsofthe14thACMSIGKDDinternationalconferenceonKnowledgediscoveryanddatamining.ACM,2008.[10]YehudaKoren,RobertBell,andChrisVolinsky.Matrixfactorization<puter,42:30C37,August2009.[11]HaoMa,DengyongZhou,ChaoLiu,MichaelR.Lyu,andIrwinKing.Recommendersystemswithsocialregularization.InWSDM’11.ProceedingsofthefourthACMinternationalconferenceonWebsearchanddatamining,pages287C196.ACM,2011.[12]MillerMcPherson,LynnS.Lovin,andJamesM.Cook.Birdsofafeather:Homophilyinsocialnetworks.AnnualReviewofSociology,27(1):415C444,2001.[13]SudeepMarwahaPunamBedi,HarmeetKaur.Trustbasedrecommendersystemforsemanticweb.InIJCAI’07.Proceedingsofthe2007InternationalJointConferenceonArti?cialIntelligence,pages.ACM,2007.[14]JasonD.M.RennieandNathanSrebro.Fastmaximummarginmatrixfactorizationforcollaborativeprediction.InICML’05.Proceedingsofthe22thInternationalConferenceonMachineLearning,pages713C719.ACM,2005.[15]BadrulSarwar,GeorgeKarypis,JosephKonstan,andJohnReidl.Item-basedcollaborative?lteringrecommendationalgorithms.InProceedingsofthe10thinternationalconferenceonWorldWideWeb,WWW’01,pages285C295,NewYork,NY,USA,2001.ACM.[16]http://en.wikipedia.org/wiki/List_of_social_networking_websites.[17]HarrisonC.White,ScottA.Boorman,andRonaldL.Breiger.Socialstructurefrommultiplenetworks.i.blockmodelsofrolesandpositions.AmericanJournalofSociology,81(4):730C780,1976.[18]ShuangHongYang,AlexSmolaH.BoLong,NarayananSadagopan,ZhaohuiZheng,andHongyuanZha.Likelikealike-jointfriendshipandinterestpropagationinsocialnetworks.InWWW’11.Proceedingsofthe20thinternationalconferenceonWorldWideWeb.ACM,2011.7.CONCLUSIONInthispaper,wedevelopajointpersonalandsociallatentfac-tor(PSLF)modelwhichcombinesthecollaborative?lteringandsocialnetworkmodelingapproachesforsocialrecommendation.Ourapproachhasshowntohavebetterpredictionaccuracyovertheexistingsocialnetworkapproaches.Inthefuturework,weplantostudythefollowingaspectsofsocialrecommendations:1)canwescalethesocialrecommendationusedthemap-reduce(Hadoop)computationframeworktohandletensorhundredsofmillionsofusers?2)howtodealwithprivacyissue?Cansocialcommendationbedonewithminimal(orcoarse)networkinfor-mation?包含各类专业文献、生活休闲娱乐、高等教育、幼儿教育、小学教育、各类资格考试、92Learning Personal Social Latent Factor Model for Social Recommendation等内容。　
　social learning theory_英语学习_外语学习_教育专区 暂无评价|0人阅读|0次下载|举报文档 social learning theory_英语学习_外语学习_教育专区。... 　a. People can change or hide personal identity. b. It’s practically impossible for them to change or hide social identity. Evidence: Body piercings →... 　需求获取模型-参考文献_计算机软件及应用_IT/计算机_...Inferring social network structure using mobile ...algorithms for the personal recommendation systems [... 　latent factor model to learn the variability of ...commerce, social networks, web search, and more....for personalized cross-domain recommendation and ... 　I. Learning English through Social Issues Suggested Items for Setting Learning Outcomes Knowledge ? ? ? ? ? ? ? Skills ? ? ? ? ? ? ? ? Values ... 　Now for me to present our point of view First of all ,I want to ...Above all,we think China should copy the western model of social ... 　latent gaussian mixture model CommunityModeling Evolutionary Behaviors for ...Recommendation Joint group and topic discovery from relations and text Social ... 　REFILE-Adidas and Nike battle for social media World Cup_计算机软件及应用_...which advises big brands on social media strategy including World Cup sponsor...