The clustering was based on the regularized SVD / matrix factorization results. Details in my e-book.