PPIs

First, use doc2vec to train the paragraph vector

then put the paragraph vector into the pair to match

then put the match results in SMOTE

and then train last