The binary model is made by word2vec:
https://github.com/dav/word2vec
Gender graph uses word embeddings trained by word2vec, the implementation is in the source repo under:
src/gender_word_plotting.c
https://github.com/sneha-belkhale/gender-word-plots
The python scripts are pretty simple. I'd like to see some methodology drill down / peer-review to make sure these models are sane. I don't really have the mathematics background myself.