#text-classification

Question 1: Single choice

You want $P(\text{spam} \mid \text{message})$ for a new SMS. Why not estimate it, or $P(\text{message} \mid \text{spam})$ , by counting how often that exact message occurs among the spam and ham messages of the training set?

Most messages are unique, so the counts would almost always be zero1
Bayes' rule cannot be applied to text data2
The counts would give probabilities that do not sum to 1 over the classes3
Spam labels are too noisy for counting to work4

#text-classification

Timed test