资源描述：
《Hongju Zhao- Efficient algorithms for segmentation of item-set time series》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、DataMinKnowlDisc(2008)17:377401DOI10.1007/s10618-008-0095-0Efficientalgorithmsforsegmentationofitem-settimeseriesParvathiChundi·DanielJ.RosenkrantzReceived:5February2008/Accepted:28March2008/Publishedonline:18April2008SpringerScience+BusinessMedia,LLC2008AbstractWep
2、roposeaspecialtypeoftimeseries,whichwecallanitem-settimeseries,tofacilitatethetemporalanalysisofsoftwareversionhistories,emaillogs,stockmarketdata,etc.Inanitem-settimeseries,eachobserveddatavalueisasetofdiscreteitems.Weformalizetheconceptofanitem-settimeseriesandpr
3、esenteffi-cientalgorithmsforsegmentingagivenitem-settimeseries.Segmentationofatimeseriespartitionsthetimeseriesintoasequenceofsegmentswhereeachsegmentisconstructedbycombiningconsecutivetimepointsofthetimeseries.Eachsegmentisassociatedwithanitemsetthatiscomputedfromt
4、heitemsetsofthetimepointsinthatsegment,usingafunctionwhichwecallameasurefunction.Wethendefineaconceptcalledthesegmentdifference,whichmeasuresthedifferencebetweentheitemsetofasegmentandtheitemsetsofthetimepointsinthatsegment.Thesegmentdifferencevaluesarerequiredtocon
5、structanoptimalsegmentationofthetimeseries.Wedescribenovelandefficientalgorithmstocomputesegmentdifferencevaluesforeachofthemeasurefunctionsdescribedinthepaper.Weoutlineadynamicprogrammingbasedschemetoconstructanoptimalsegmentationofthegivenitem-settimeseries.Weuset
6、heitem-settimeseriessegmentationtechniquestoanalyzethetemporalcontentofthreedifferentdatasetsEnronemail,stockmarketdata,andasyntheticdataset.Theexperimentalresultsshowthatanoptimalsegmentationofitem-settimeseriesdatacapturesmuchmoretemporalcontentthanasegmentationc
7、onstructedbasedonResponsibleeditor:EamonnKeogh.P.Chundi(B)ComputerScienceDepartment,UniversityofNebraskaatOmaha,Omaha,NE68106,USAe-mail:pchundi@mail.unomaha.eduD.J.RosenkrantzComputerScienceDepartment,SUNYatAlbany,Albany,NY12222,USAe-mail:djr@cs.albany.edu123378P.C
8、hundi,D.J.Rosenkrantzthenumberoftimepointsineachsegment,withoutexaminingtheitemsetdataatthetimepoints,andcanbeusedtoanalyzedifferenttypesoftempor
