Effect of Choosing Loss Function when Using T-batching for Representation Learning on Dynamic Networks