Rank Loss for News feed / Recommendation problem

How do I use the rank loss (or triplet loss) for a feed ranking or recommendation problem? In other words, how do we create the training data for rank loss (or triplet loss) model training? Thanks!