topic model power law pitman yor process issei sato hiroshi nakagawa import approach knowledg discoveri data mine estim unobserv variabl latent variabl hidden specif properti observ data latent factor model assum item record latent factor occurr item model latent factor document model record document repres bag word mean order word item word latent factor topic latent dirichlet alloc lda wide bayesian topic model appli dirichlet distribut latent topic distribut document multipl topic lda assum latent topic discret latent variabl distribut multinomi distribut paramet gener dirichlet distribut lda model word distribut multinomi distribut paramet dirichlet distribut dirichlet multinomi set captur power law phenomenon word distribut zipf law linguist propos topic model pitman yor py process call py topic model py topic model captur properti document power law word distribut presenc multipl topic experi real data model outperform lda document model term perplex 