Naïve Bayes for text classification

It is perhaps surprising that an algorithm based on calculating conditional probabilities could be useful for text classification. But this follows fairly straightforwardly with a key simplifying assumption. Let’s assume that our documents can be well represented by the counts of each word in the document, without regard for word order or grammar. This is known as a bag-of-words. The relationship that a bag-of-words has to a categorical target – say, spam/not spam or positive/negative – can be modeled successfully with multinomial naïve Bayes.  

We will work with text message data in this section. The dataset we will use contains labels for spam and not spam messages.

Note

This dataset on text messages can be downloaded by the public at https://www.kaggle.com/datasets/team-ai/spam-text-message-classification. It contains two columns: the text message and the spam or not spam (ham) label.

Let’...