Can anyone here provide an intuitive explanation for why you'd prefer NMF over SVD or Truncated SVD? It seems to be simply another way of accomplishing somethin

Can anyone here provide an intuitive explanation for why you'd prefer NMF over SVD or Truncated SVD? It seems to be simply another way of accomplishing something very similar, but with this non-negativity constraint. But it's not obvious to me why that would be useful, or produce a better factorization. Can anyone shed any light on that? My sense from reading about it and understanding what's going on to the extent that I do, is that it's simply an alternative factorization. And there's no real reason to prefer one to the other, but since they are similar but non-equivalent you can try both and see which one works better.