Why ML.NET Matrix Factorization makes no sense predictions

Question

I developed a product recommender using ML.NET matrix factorization. The train data has 20000 entries and contains only pairs of product ids (no ratings), so I adapted the t