Comment by BiasRegularizer

Nice job OP. I noticed that you used word2vec for the embedding space, do you think more modern ones like seq2seq or BERT might give better performance?