Latent Dirichlet Allocation Paper

Observing the patterns of patients, in order to detect the groups of patients, we design a generative process of the patterns incorporating diagnosis groups as latent variables.
First, we assume the patterns in a particular diagnosis group follow a multinomial distribution, which is a random variable draws from a prior Dirichlet distribution.
By making analogies between words in documents and patterns of patients and inspiring from bag-of-words, we represent the observed patterns of a patient as a bag of patterns. Therefore, the generative process can be treated as the process of Latent Dirichlet Allocation (LDA)~\cite{blei2003LDA}.
More specifically, patterns of a patient are generated as: (1) A patient has $N$ patterns to describe his/her