← AI Terminology

Word2Vec

Word2Vec learns dense vector representations of words from local context (CBOW or skip-gram) so algebraic relationships emerge in embedding space.

Mikolov et al.’s method ignited the modern embedding era.
Why It Matters in AI
Word2Vec proved distributional semantics could be learned at scale with simple objectives — paving the way for GloVe, fastText, and eventually contextual models like BERT. Still useful pedagogically and for lightweight features.
Key Points
Aspect Description
Use Init features, educational demos, some production relics
Limit One vector per word type; no context sense
Famous king − man + woman ≈ queen style analogies
Related Embedding, contrastive learning lineage
Algorithms CBOW and skip-gram with negative sampling
Successors GloVe, fastText, contextual transformers
Simple Analogy
Placing words on a map so those that share neighbours in sentences end up as geographic neighbours — “coffee” near “tea,” far from “hyperbola.”
Common Usage Examples
  • gensim.models.Word2Vec
  • Analogy evaluations
  • Average word vectors for doc features
  • Contrast with BERT contextual vectors
Summary
In short: Word2Vec learns static word embeddings from context — the breakthrough that launched modern representation learning for text.