Abstract:

We propose and use a privacypreserving Naive Bayes classifier to the issue of private text classification. In this scenario, one side (Alice) is holding a text message, while another (Bob) is holding a classifier. Alice will only learn the outcome of the classifier applied to her text input at the conclusion of the protocol, whereas Bob will learn nothing. Secure Multiparty Computation is the foundation of our solution (SMC). Our Rust implementation offers a quick and safe solution for unstructured text categorization. In the event when Bob's model's dictionary size covers all words (n = 5200) and Alice's SMS comprises at most m = 160 unigrams, we can identify an SMS as spam or ham in less than 340 ms (the solution is general and may be applied in any other scenario in which the Naive Bayes classifier can be utilised). Our method takes just 21 ms for n = 369 and m = 8 (the average of a spam SMS in the database).

Description:

Privacy-Preserving Classification, Secure Multiparty Computation, Naive Bayes, Spam.