Achieves 0.55 MRR (and 0.64 MRR on the specific subset of data) at the RecSys2019 challenge using a two stage deep neural network architecture. More information can be gathered from the Documentation.pdf in particular at the "Deep Learning Model" section..
Eventuell wäre es sinnvoll einfach Random Rows zu duplizieren.