model base document cluster collaps gibb sampler daniel david walker eric ringger model base algorithm emerg prefer method document cluster comput resourc improv method gibb sampl common paramet estim model gibb sampl understood applic extens studi document cluster explor converg rate possibl label switch chain summar methodolog document cluster model mixtur multinomi model show fairli simpl method emploi produc cluster superior qualiti compar produc em algorithm 