confused about "SampledSoftmaxLoss" func

[zhhu1996]

Hey, Congratulations for your perfect and creative work. when I read the implementation code here, I am very confused about SampledSoftmaxLoss. I have some questions for this:

1. why do we use "supervision_ids" to calculate "positive_logits"？
2. why wu use "InBatchNegativesSampler" to random sample negative samples and calculate "negative_logits"?
3. what does the "self._model.interaction" do?
4. for jaggled_loss, why need to firstly concat in 1 dim and then calculate log_softmax in 1 dim and last pick the 0 dim?

Please give me some advice if you are free, thanks~