In this paper we propose a Dirichlet process topic model for representing patterns in complex dynamic data. We assume that in a dynamic corpus topics do not randomly appear and disappear but rather replace each other following transitional rules. We model topic transitions as a Markov chain with an infinite hidden Markov model. The proposed model is then formulated as a Dirichlet process with a hierarchical Dirichlet process hidden Markov model prior.
|Publication status||Published - 2018|
|Event||All of Bayesian Nonparametrics (Especially the Useful Bits): A workshop at the Thirty-Second Annual Conference on Neural Information Processing Systems (NeurIPS 2018). - Palais des Congrès de Montréal, Montréal, Canada|
Duration: 7 Dec 2018 → 7 Dec 2018
|Workshop||All of Bayesian Nonparametrics (Especially the Useful Bits)|
|Abbreviated title||BNP@NeurIPS 2018|
|Period||7/12/18 → 7/12/18|