What is feature weighting in filtering systems?

Feature weighting in filtering systems assigns different importance to various signals based on their predictive value. Authentication status might carry high weight because authenticated senders are rarely spammers. Certain content patterns might have lower weight because they appear in both spam and legitimate mail.

Weights are learned from training data, not manually assigned. The model discovers which features best predict spam versus legitimate mail and adjusts weights accordingly. This data driven approach finds patterns humans might not anticipate.

Weights can vary by context. A feature strongly predictive of spam in consumer email might be less relevant for business email. Sophisticated systems learn context specific weights, applying different models or weight adjustments based on recipient characteristics or message categories.