An Improved Algorithm for Training Field-aware Factorization Machines (FFM)
Existing FFM training algorithms are not efficient when some fields contain a huge number of features because they may calculate the pair-wise interactions implicitly. The proposed method reduces the complexity from square to linear of feature numebr.
Assume that you're using Linux or Cygwin with all necessary Latex packages installed. You can simply type
make
in your command line.