Furthermore, in the cases of the variational approximation scheme, the topic model method cannot be even trained as the number of latent topics increases. However, reasonable results are obtained with the Gibbs sampling scheme. We argue that this is be