Generallyadatabaseencompassesvariouskindsofknowledgeandissharedbymanyusers.Differentusersmaypreferdifferentkindsofknowledge.Soitisimportantforadataminingalgorithmtooutputspecificknowledgeaccordingtousers'currentrequire-ments(preference).Wecallthiskindofdataminingrequirement-orientedknowledgediscovery(ROKD).Whentheroughsettheoryisusedindatamining,theROKDproblemishowtofindareductandcorrespondingrulesinterestingfortheuser.Sincereductsandrulesaregeneratedinthesameway,thispaperonlyconcernswithhowtofindaparticularreduct.Theuser'srequirementisdescribedbyanorderofattributes,calledattributeorder,whichimpliestheim-portanceofattributesfortheuser.Intheorder,moreimportantattributesarelocatedbeforelessimportantones.Thentheproblembecomeshowtofindareductincludingthoseattributesanteriorintheattributeorder.Anapproachtodealingwithsuchaproblemisproposed.Anditscompletenessforreductisproved.Afterthat,threekindsofattributeorderaredevelopedtodescribevarioususerrequirements.
