Word Mover's Distance

Built on Word2Vec, WMD regards a document as the collection of word vectors prescribed Word2Vec, and minimises the total travel distance of moving all vectors in one document to another.

The minimised travel distance is used as a measure of dissimilarity of the two documents - the higher the WMD, the more dissimilar the pair of documents.

WMD coupled with Google's Word2Vec does not require pretraining after some time, unlike most other text analysis models which require training first on entire corpus to extract latent topics and retraining to capture changes in corpus.