[minimaxir]

What may be interesting is to use cosine similarity between the embeddings of these words to see if synonyms can be accurately identified.

Awhile ago, SpaCy set up a demo doing just that on the Reddit dataset:

https://demos.explosion.ai/sense2vec/?word=cannabis&sense=au...

https://demos.explosion.ai/sense2vec/?word=marijuana&sense=a...