Multi-media Fusion through Application of Machine Learning and NLP

《Multi-media Fusion through Application of Machine Learning and NLP》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、Multi-MediaFusionthroughApplicationofMachineLearningandNLPChinatsuAone,ScottWilliamBennettandJimGorlinskySystemsResearchandApplicationsCorporation(SRA)4300FairLakesCourtSouthBuildingFairfax,VA22033-4232aonec@sra.com,bennett@sra.com,gorlinsk@sra.comAbstractThispaperdescribeswor

2、konasystemtodynam-icallyclusterandpresentinformationfromlivemultimedianewssources.Featuresareobtainedbyapplyingstatisticalandnaturallanguagepro-cessingtechniquestotextsassociatedwiththemedia(newwirestoriesorclosed-captions).Con-ceptualclusteringisemployedtodynamicallybuildaset

3、ofhierarchicalclusters.Resultsarepresentedobjectivelycomparingautomatedclus-teringwithhandclusteringofthesamearticles.IntroductionWiththeever-increasingamountofinformationavail-abletodayinprintedandelectronicform,itisbecom-ingmoreandmoredicultforpeopletoassimilateFigure1:Syst

4、emsnapshotinformationfromalltheavailablesources.Inthispa-per,wewilldiscussaprototypesystemfortheMulti-MediaFusion(MMF)projectwhichenablestheauto-prolesselectedinadvance.Therefore,theMMFsys-matedassimilationofsuchinformation.Endusersoftemmustbedierentfromroutingsystemswhichar

5、ethissystemwouldgreatlybenetfromthespeedandtrainedbyknowingrelevantandnon-relevantsetsofeaseofaccesstorelevantinformation.Theinforma-articles.tioncollectedcanconsistoftext,video,sound,and/orCurrentlythesystemusesCNNHeadlineNewsandphotos.TheMMFsystemcanautomaticallyfuseCNNtheC

6、lariNetnewsfeedasmultiplesources.Wede-HeadlineNewsclipswithrelatednewsarticlesusingacodeandutilizetheclosed-captiontextassociatedwithconceptualclusteringalgorithmbasedonCobwebandCNNHeadlineNewsinordertofusethevideoclipshigh-levelfeaturesderivedbyanaturallanguagesys-withrelated

7、newsarticles.TheCNNclosed-captiontem.Theresultisamultimediadocumentwherein-textissegmentedintodierentnewstopicsaccordingformation,whethertextorvideo,onrelatedeventsistotheclosed-captionformat.Thevideo,audio,andgroupedtogether.Thus,theusercanquicklyassimi-captionscanbetransmit

8、tedlivewhileclipsaresimul-lateinformationofinterestbychoosing