What is feature likelihood estimation in Multinomial Naïve Bayes with smoothing?

Likelihood is estimated as smoothed normalized word counts per class.

In Multinomial NB, likelihood P(w|C) is computed as (count(w,C)+α)/(total_words_in_class + α*V), where V is vocabulary size. Smoothing prevents zero probabilities and stabilizes rare word effects. This is crucial in NLP where unseen words are common.