To combat overly long, keyword stuffed documents (hello 2,000 word articles my old friend), information retrieval systems have a Pivoted Document Length Normalisation mechanic.

Whilst still measuring the frequency a word appears (Term Frequency) and query-document similarity scores (Euclidean Length), term and topic bloat is countered by normalisation.

Think of it like a golf handicap.

‘You sure can drive the ball far, but your short game is shit.’