Demo of word2vec in Gensim https://radimrehurek.com/gensim/models/word2vec.html for shallow-learning of word similarities based on context. This uses 3,000,000 words in the vocabulary using Google's pre-built model.
A targetted training run should be performed to improve accuracy.