RCLens Interactive Rare Category Exploration and Identification.[2017][IEEE Trans Vis Comput Graph][10.1109tvcg.2017.2711030]

《RCLens Interactive Rare Category Exploration and Identification.[2017][IEEE Trans Vis Comput Graph][10.1109tvcg.2017.2711030]》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、Thisarticlehasbeenacceptedforpublicationinafutureissueofthisjournal,buthasnotbeenfullyedited.Contentmaychangepriortofinalpublication.Citationinformation:DOI10.1109/TVCG.2017.2711030,IEEETransactionsonVisualizationandComputerGraphics1RCLens:InteractiveRareCategoryExplorationandIdentificationHanfe

2、iLin,SiyuanGao,DavidGotz,FanDu,JingruiHe,andNanCaoAbstract—Rarecategoryidentificationisanimportanttaskinmanyapplicationdomains,rangingfromnetworksecurity,tofinancialfrauddetection,topersonalizedmedicine.Theseareallapplicationswhichrequirethediscoveryandcharacterizationofsetsofrarebutstructurally-

3、similardataentitieswhichareobscuredwithinalargerbutstructurallydifferentdataset.ThispaperintroducesRCLens,avisualanalyticssystemdesignedtosupportuser-guidedrarecategoryexplorationandidentification.RCLensadoptsanovelactivelearning-basedalgorithmtoiterativelyidentifymoreaccuraterarecategoriesinres

4、ponsetouser-providedfeedback.Thealgorithmistightlyintegratedwithaninteractivevisualization-basedinterfacewhichsupportsanovelandeffectiveworkflowforrarecategoryidentification.Thispaper(1)definesRCLens’underlyingactive-learningalgorithm;(2)describesthevisualizationandinteractiondesigns,includingadis

5、cussionofhowthedesignssupportuser-guidedrarecategoryidentification;and(3)presentsresultsfromanevaluationdemonstratingRCLens’abilitytosupporttherarecategoryidentificationprocess.IndexTerms—Visualanalytics,informationvisualization,rarecategorydetection,machinelearning.F1INTRODUCTIONVARIETYofalgorit

6、hmshavebeendevelopedtoidentifyWhilewidelyuseful,rarecategorydetectionisatechnicallyApredominanttrendsorclassesofobjectswithincomplexchallengingtaskthatiscurrentlyanactiveresearchtopicwithindatasets,includingregression,classification,andclustering.Thesethemachinelearningcommunity(e.g.[12,13]).One

7、factortechniquesaimtocharacterizethelarge-scalestructureswithinmakingthistasksodifficultisthelackofknowncategoriesadatasetthathavehighlevelsofsupport.Thisapproachhasainmostreal-worldapplications.Ratherthanseekingtoidentifywidesprea