What is TF-ISF?
TF-ISF scores how important a word is to a sentence by combining how often it appears in that sentence with how rare it is across all the sentences of the document. Sentences full of distinctive words score highest.
TF-ISF formula
For a term t in a sentence s, in a document of N sentences:
TF-ISF(t, s) = tf(t, s) × log(N ÷ n(t))
Here tf(t, s) is how often t appears in s, and n(t) is how many sentences contain t. A sentence’s score is usually the sum or average of its terms’ scores, after very common words are removed.
A TF-ISF example
In a 40-sentence article, “pricing” appears in 20 sentences and “annual” in 2. Using natural logarithms, the rarer word gets the higher weight: log(40 ÷ 2) ≈ 3.0 against log(40 ÷ 20) ≈ 0.7. A sentence about the annual plan therefore outscores one that only mentions pricing.
TF-ISF vs TF-IDF
TF-IDF does the same job across a collection of documents: inverse document frequency measures how rare a word is across many documents. TF-ISF moves the idea inside one document, treating each sentence as a small document of its own, which is what you need to decide which sentences of an article matter most.
Where TF-ISF is used
TF-ISF sentence scoring is a classic method for extractive summarization: score every sentence, keep the best. AI Powered Content Repurposing ranks sentences this way, so the posts and quotes it pulls out are the most distinctive lines in your source.
Last reviewed 2026-11-01.