Assessing ranking metrics in top-N recommendation