基于微博的知识词条推荐算法-研究

《基于微博的知识词条推荐算法-研究》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。

1、哈尔滨工业大学工学硕士学位论文AbstractWiththedevelopmentoftheInternet,thewayofoursociallifeandinformationacquiringhaschangeddramatically.Theriseofmicroblogsallowspeopletorapidlyaccesstheenormousamountofinformation,soitisimportanttodiscoverusefulinformationautomaticallyandrecom

2、mendtousers.Takingadvantageofknowledgediscoverymethodstodiscoverusefulinformationinmassivedataandusingsocialnetworkstosolvethesparsenessprobleminthetraditionalrecommendationalgorithmsareissuesofcurrentresearch.Microblog-basedknowledgediscoveryandrecommendationar

3、eproposedagainstthebackgroundofbigdataandpersonalizedera.Extractingknowledgefrommassivemicroblogsandrecommendingtheseknowledgetouserswhomightbeinterestedinthemarekeypointsofthisresearch.Whenconstructingthetrainingcorpus,wefindthatmostoftheknowledgeentrydiscovery

4、corpusesareconstructedfromthestandardtextsandtherehaven’tbeenanyopenstandardcorpusconstructedfrommicroblogs.Atthemeantime,socialnetwork-basedcorpusesaremainlyusedtorecommendmusicandfriends,andwecan’tfindanystandardcorpusforrecommendingmicroblogknowledgeentries.T

5、herefore,inthispaperweconstructacorpusformicroblogknowledgeentrydiscoveryandacorpusformicroblogknowledgeentryrecommendationbyusingamicroblogcrawlertogatherdataofmicroblogsandsocialrelationshipbetweenusers.Inthemicroblogknowledgeentrydiscoverytask,aCRFsmodelisado

6、ptedtoidentifyknowledgeentriesinthecorpus.InordertoimprovetherecallofthetraditionalCRFsmodel,wetrainwordclusteringsonunlabeleddataandconstructadictionaryfromthetrainingcorpus,thenmergethemintotheCRFsmodel.Asaresult,F1scoreofCRFsmodelwithwordclusteringfeaturesout

7、performstheonewithbasicfeaturesatanimprovementof0.0656,andtheCRFsmodel’swithdictionaryfeatureraisesby0.0805,andCRFsmodel’swithbothfeaturesraisesby0.0843.Besides,wecomparetheeffectofdifferentnumberofclustersanddifferentsizeofcorpushasonwordembeddingfeatures.Inthe

8、microblogknowledgeentryrecommendationtask,weusesocialrelationshipsbetweenusersandtimefactortoimprovethetraditionalcollaborativefilteringalgorithms,andcompareitwiththe