How does Naïve Bayes scale with extremely large vocabularies in NLP systems?

Naïve Bayes scales linearly with vocabulary size using sparse representations and efficient counting.

In NLP systems, vocabularies can exceed millions of tokens. NB handles this efficiently using sparse matrices and storing only non-zero counts. Training complexity is O(N * V_nonzero), and inference ignores zero-probability features. Hashing tricks and feature pruning further improve scalability.