Semi-supervised graph ranking with rich meta data

《Semi-supervised graph ranking with rich meta data》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、Semi-SupervisedRankingonVeryLargeGraphwithRichMetadata∗BinGaoTie-YanLiuWeiWeiMicrosoftResearchAsiaMicrosoftResearchAsiaComputerSoftwareand4F,SigmaCenter,No.49,4F,SigmaCenter,No.49,TheoryZhichunRoadZhichunRoadHuazhongUniversityofBeijing,100190,P.R.China

2、Beijing,100190,P.R.ChinaScienceandTechnologybingao@microsoft.comtyliu@microsoft.comWuhan,430074,P.R.Chinaweiwei8329@gmail.comTaifengWangHangLiMicrosoftResearchAsiaMicrosoftResearchAsia4F,SigmaCenter,No.49,4F,SigmaCenter,No.49,ZhichunRoadZhichunRoadBeij

3、ing,100190,P.R.ChinaBeijing,100190,P.R.Chinataifengw@microsoft.comhangli@microsoft.comABSTRACTthenodes.Finally,weshowthatitispossibletomakethealgorithmefficienttohandlebillion-nodegraphbytakingGraphrankingplaysanimportantroleinmanyapplica-advantageofthes

4、parsityofthegraphandimplementitintions,suchaspagerankingonwebgraphandentityrank-theMapReducelogic.Experimentsontherealdatafromaingonsocialnetworks.Intheapplications,besidesgraphcommercialsearchengineshowthattheproposedalgorithmstructure,richinformation

5、onnodesandedgesandexplicitcanoutperformpreviousalgorithmsonseveraltasks.orimplicithumansupervisionareoftenavailable.Incon-trast,conventionalalgorithms(e.g.,PageRankandHITS)computerankingscoresbyonlyresortingtographstructureCategoriesandSubjectDescripto

6、rsinformation.Anaturalquestionariseshere,thatis,howH.3.3[InformationStorageandRetrieval]:InformationtoeffectivelyandefficientlyleveragealltheinformationtoSearchandRetrieval;H.5.4[InformationInterfaceandmoreaccuratelycalculategraphrankingscoresthanthecon-P

7、resentation]:Hypertext/Hypermedia.ventionalalgorithms,assumingthatthegraphisalsoverylarge.Previousworkonlypartiallytackledtheproblem,andtheproposedsolutionsarealsofarfrombeingsatisfactory.GeneralTermsThispaperaddressestheproblemandproposesageneralAlgor

8、ithm,Experimentation,Theoryframeworkaswellasanefficientalgorithmforgraphrank-ing.Specifically,wedefineasemi-supervisedlearningframe-workforrankingofnodesonaverylargegraphandderiveKeywordswithinourproposedframeworkanefficientalgorithmcalledPag