For LDA, this would be equivalent to a model with one V-dimensional vector of word-topic probabilities (\beta) governing word probabilities: f(\beta | w) \propto f(w | \beta) f(\beta | \gamma)
For SLDA, we would have an intercept-only model for y: y_d ~ N(\eta_0, \sigma^2)
For SLDAX, we would have y_d ~ N(\eta_0 + x_d' \eta_x, \sigma^2)
For LDA, this would be equivalent to a model with one V-dimensional vector of word-topic probabilities (\beta) governing word probabilities: f(\beta | w) \propto f(w | \beta) f(\beta | \gamma)
For SLDA, we would have an intercept-only model for y: y_d ~ N(\eta_0, \sigma^2)
For SLDAX, we would have y_d ~ N(\eta_0 + x_d' \eta_x, \sigma^2)