document cluster dirichlet process mixtur model featur select guan yu ruizhang huang zhaojun wang essenti issu document cluster estim number cluster document collect document partit paper propos approach dpmf address issu propos approach design group document set cluster number document cluster determin dirichlet process mixtur model automat identifi discrimin word separ irrelev nois word stochast search variabl select techniqu explor perform propos approach synthet dataset realist document dataset comparison propos approach stage art document cluster approach approach robust effect document cluster 