gensim may be slower than what we can do manually, perhaps with some help from nltk. Using gensim, we recalculate pairings of texts, but since cosine similarity commutes, we only need to calculate one triangle of the resulting matrix of results.
Giorgio and I compared the results: the similarity scores produced by the old custom calculations and the new gensim calculations are very close, but the gensim code is faster.
gensim may be slower than what we can do manually, perhaps with some help from nltk. Using gensim, we recalculate pairings of texts, but since cosine similarity commutes, we only need to calculate one triangle of the resulting matrix of results.