What is the role of prior smoothing in Bayesian regularization of Naïve Bayes?

Prior smoothing acts as Bayesian regularization to prevent overfitting in sparse datasets.

Smoothing techniques like Laplace or Lidstone smoothing introduce pseudo-counts into probability estimation. This corresponds to placing a Dirichlet prior over categorical distributions. It prevents zero probabilities and reduces variance in small-sample regimes, improving generalization.