featur select method text classif anirban dasgupta petro drinea boulo harb vanja josifovski michael mahonei featur select text classif theoret empir main result unsupervis featur select strategi give worst case theoret guarante gener power result classif function respect classif function obtain keep featur knowledg featur select method guarante addit analysi lead insight featur select strategi perform practic techtc newsgroup reuter rcv data set evalu empir perform simpler relat featur select strategi commonli strategi empir evalu show strategi provabl perform guarante perform comparison commonli featur select strategi addit perform dataset aggress featur select 