cluster base classif domain document wenyuan dai gui rong xue qiang yang yong yu real world applic label data short suppli obtain label data domain expens time consum plenti label data relat domain tradit machin learn cope learn domain paper address problem text mine task label data distribut domain domain data unlabel data relat domain domain data gener goal learn domain appli learn knowledg domain propos cluster base classif cocc algorithm tackl problem cluster bridg propag class structur knowledg domain domain present theoret empir analysi show algorithm produc high qualiti classif result distribut data experiment result show algorithm greatli improv classif perform tradit learn algorithm 