Why ML.NET MatrixFactorization is getting different results after each training and outputs no sense recommendations

Question

I developed a product recommender using ML.NET matrix factorization. The train data contains only pairs of product ids, so I adapted the trainer accordingly to Microsoft doc