Word2vec was the Google technique introduced in 2013 for learning vector representations of words from large text collections.
The method represents words as numerical vectors, allowing a computer to compare relationships in a continuous mathematical space. Words that occur in similar contexts tend to receive vectors that are close together, helping models capture aspects of meaning and usage.
Word2vec used shallow neural-network architectures, especially continuous bag-of-words and skip-gram models. Its efficient training made it practical to learn representations from very large corpora, and its release helped popularize word embeddings across natural-language processing.
Word2vec is not a dictionary or a language model that writes complete passages. It is primarily a method for learning word representations. GloVe and FastText are related alternatives, while WordNet is a manually organized lexical database rather than a neural embedding technique.