Which machine learning paradigm is most commonly used to build spam filters that improve over time based on user feedback (marking emails as spam or not spam)?

Why B is correct

Spam filters retrain on labeled examples: each email marked spam or not-spam by the user becomes a labeled training pair for a supervised classifier. This is the defining characteristic of supervised learning. Unsupervised clustering can discover patterns but cannot directly optimize for the spam/not-spam boundary without labels. RL could theoretically be used but is not how practical spam filters work; user opens are not the right reward signal and the feedback loop is too sparse. Self-supervised learning is used for language model pretraining, not for binary classification with explicit user labels.