The Netflix prize has rejuvenated a widespread interest in the matrix factorization approach for collaborative filtering. We describe a simple algorithm for incorporating content information directly into this approach. We present experimental evidence using recipe data to show that this not only improves recommendation accuracy but also provides useful insights about the contents themselves that are otherwise unavailable.