Figure 9. Latent Dirichlet Allocation model.

The nodes represent random variables. The shaded nodes are observed variables and the unshaded nodes indicate unobserved variables. The plates indicate repetitions. M is the number of video sequences of each action, N_d is the number of code words w in the d-th video sequence, and z is the hidden topic. The parameters of the model include α, the parameter of the Dirichlet distribution, θ, the parameter of the multinomial distribution, and β, which parameterizes the multinomial distribution conditioned on topic z.