Table 3 Procedure of the General MFNN algorithm

Algorithm 1: The General MFNN Algorithm

Input:Y: the known interaction matrix

Set: Epoch: e, Batch size: b, Learning rate: l

Output:W: model parameters

1: Randomly sample the train set Y^train  and validation set  Y^vali from  Y.

2: Initialize the model parameters  w_{c _ n}  and  w_{r _ n}  with a Gaussian distribution

3: while not model is converged and epoch > e do

 sample a mini batch from  Y^train  in size b

 set p_ri  and  q_cj  using Eq. 2 and 3 with mini batch

 set  f_{i, j}  using Equation 4 with  p_ri  and  q_cj

 set  L using Equation 5 with  f_{i, j} and  y_ij

 use Adam optimizer to optimize model parameters

end while

4: using the  Y^vali  evaluate the model