← AI Terminology

TF-IDF - Term Frequency–Inverse Document Frequency

TF-IDF weights terms by how often they appear in a document (TF) discounted by how common they are across the corpus (IDF) — a classic sparse text representation.

Still strong for keyword retrieval and linear baselines.
Why It Matters in AI
Before dense embeddings dominated, TF-IDF vectors powered search and text classification. It remains a robust baseline and a building block of hybrid retrieval with BM25-related ideas.
Key Points
Aspect Description
TF Term frequency in a document
IDF Log-scaled rarity across documents
Use Search features, linear SVM/logreg text models
Tools sklearn TfidfVectorizer
Limits No true semantics/synonyms
Related BM25, bag-of-words, embeddings
Simple Analogy
Highlighting words that show up a lot in this article but rarely in the rest of the library — distinctive keywords glow.
Common Usage Examples
  • TfidfVectorizer + logistic regression baseline
  • Keyword features for ranking
  • Compare to embedding classifiers
  • Sparse vectors for interpretable models
Summary
In short: TF-IDF scores distinctive keywords in documents — the classic sparse text feature that still baselines many NLP systems.