ความถี่ของคำและความถี่ผกผันของเอกสาร

TF-IDF is built from two pieces that multiply together: term frequency and inverse document frequency. Each half fixes a different weakness of raw counts.

Term frequency measures how often a word appears inside one document. More mentions of a word suggest that document leans toward that topic.

We often divide a word's count by the document length. This normalization stops long documents from looking important just because they have more words.

By itself, term frequency still rewards filler words that appear a lot. We need a second factor to punish words that show up everywhere.

Document frequency counts how many documents contain a word at all. A high document frequency means the word is common across your whole collection.

We want rare words to score high, so we invert that count. Inverse document frequency rises when a word appears in few documents and falls when it is everywhere.

IDF uses a logarithm so the values do not explode for very rare words. The log keeps the scale smooth and comparable across terms.

The final score is simply TF times IDF. A word wins only if it is frequent here and rare elsewhere, which is exactly the combination we wanted.

A topic word like neuroscience in one article gets a high score. A word like the gets crushed because its IDF is nearly zero.

Words appearing in almost every document earn tiny weights. TF-IDF automatically demotes this background noise without any manual stopword list.

TF-IDF balances local importance against global rarity in one clean formula. That balance is why this weighting stayed a search and text staple for decades. ⭐

What does the inverse document frequency part actually reward?

TF measures local frequency, IDF rewards global rarity, and their product is TF-IDF. Together they spotlight words that truly define a document. ✅