enhanc semi supervis cluster featur project perspect wei tang hui xiong shi zhong jie wu semi supervis cluster emploi limit supervis form label instanc pairwis instanc constraint aid unsupervis cluster significantli improv cluster perform vast amount expert knowledg spent problem exist work design handl high dimension spars data paper fill crucial void develop emi supervis cluster method base sphe ical an atur projectio screen specif formul problem constraint guid featur project nice integr semi supervis cluster algorithm abil effect reduc data dimens experiment result real world data set show screen method effect deal high dimension data appeal cluster perform 