categori detect hierarch shift pavan vatturi weng keen wong applic surveil monitor scientif discoveri data clean requir identif anomali method develop identifi statist signific anomali difficult task identifi anomali interest statist signific categori detect emerg area machin learn address issu human loop approach interact set algorithm ask user label queri data point exist categori declar queri data point belong previous undiscov categori goal categori detect bring user attent repres data point categori data queri data set imbalanc categori main challeng identifi rare categori anomali task refer rare categori detect present approach rare categori detect base hierarch shift approach hierarchi creat repeatedli appli shift increas bandwidth data hierarchi identifi anomali data set scale pose queri user main advantag methodolog exist approach requir knowledg dataset properti total number categori prior probabl categori result real world data set show hierarch shift approach perform consist previou techniqu 