By　using　the　non-linear　operators　(such　as　Isomap,　WordNet　etc.),　how　to　promote　the　traditional　text　information　processing　techniques　to　＇understanding＇　level　was　discussed.　Based　on　manifold　analysis,　the　distribution　of　Chinese　words　in　a　continuous　semantic　space　was　primarily　studied.　Short　messages　clustering　based　on　WordNet　was　investigated.　Experimental　results　prove　that　these　methods　can　reflect　the　internal　relation　of　texts.