Natural Language Processing and Computational Linguistics

Poincare

The last word embedding technique we will look at is Poincare embeddings, also developed by the good folks over at Facebook AI research. The general idea is to use a graphical representation of words to better understand the relationship between words and to generate the word embeddings. Poincare embeddings can also capture hierarchical information using this graphical representation – in the original paper [44], titled Poincaré Embeddings for Learning Hierarchical Representations, this hierarchical information is learned by using WordNet noun hierarchy. This information is calculated in the hyperbolic space, and not the traditional euclidean space – allowing for us to better capture the notions of hierarchy.

Gensim's notebook directory ...