基于微博的热点话题发现

《基于微博的热点话题发现》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、AbstractHotTopicExtractionfromMicroblogsMajor：ComputerApplicationAuthor：YingZhuSupervisor：Prof．LiLiAbstractWiththedevelopmentofinformationtechnology,theInternetdataandresourcesincreasemassively．Inordertoeffectivelymanageandutilizethiskindofinformatio

2、n,content-basedinformationretrievalanddatamininghavebecomehotresearchtopics．Latentsemanticanalysis，whichcomputestopicdistributionforthedocumentandworddistributionforthetopicbasedonknownworddistributionforthedocumentinordertogetlatenttopics，playsanimp

3、ortantroleininformationretrievalandtextminingandisappliedextensivelyinthefieldoftextclassificationandclustering，informationorganizationandmanagementandhottopicextraction．Inrecentyears，withtheriseofWeb2．0，socialnetworks，suchasRenren,Facebook，Twitter,S

4、inaWeibo，etc．notonlybecomeverypopular,butalsobecomeawayofmodemlife．Widelyusedsocialmediaproducesmassiveuser-generatedcontentdata(UGC)，ofwhichmorethan80％isnaturallanguagetext．Itisobservedthatdetectinghottopicscarlbeveryhelpfulandnecessaryforpeopletoge

5、tessentialinformationquickly．However,thesetextsarespecialandbeartheirowncharacteristics，manytraditionaltopicanalysismodelcannotachievebetterresultsunlessaugmentedwithnewfeatures．Textsfromsocialnetworkshavefoursalient：high-dimensional，sparse，notnormat

6、iveandunevendistributionoftopics．Inotherwords，largenumbersofmessagesarepostedineveryminuteandthesetextslikelyproducethevectorwithmorethantenthousandsofdimensions，whichistootime-consumingforthetopicextraction；comparedwiththelongtexts，thesetextshaveeve

7、nlesskeywords，producingthesparse“document-word'’matrixandthusdifficulttoextracttheeffectivefeaturesandtoexploitthecorrelationbetweenthefeatures；abbreviationsandcatchwordsareusedextensivelyinsocialnetwork，increasingthesynonymsinthetextsandmakingthetop

8、icidentificationtaskmoredifficulttohandle；inaddition，fewermessagesonmicroblogsarevaluableforhottopicdetectionduetomassiveisaboutusers’dailylife，suchasweather,foods，emotions，andSOon．SowhetheratermishotornotisnotbasedonitsfrequencyofoCCUrn翻nCe．III西南大学硕