hierarch mixtur model probabilist analysi mark sandler mixtur model form wide class gener model describ structur cluster data paper develop approach analysi hierarch mixtur model specif text cluster problem motiv describ natur gener process creat hierarch mixtur model data process adversari start arbitrari base distribut build topic hierarchi evolutionari process control paramet process prove assumpt subset topic repres gener basebal sport base document produc topic hierarchi effici determin special topic subset belong qualiti classif independ total number topic hierarchi algorithm total number topic advanc approach yield algorithm cluster unsupervis topic tree reconstruct valid model show properti predict theoret result carri real data appli cluster algorithm dataset newsgroup ii snapshot abstract arxiv categori abstract case algorithm perform extrem 